阅读说明：本技术 使用机器学习的组合的室内和室外跟踪 (Combined indoor and outdoor tracking using machine learning ) 是由 尼古拉斯·布特韦格 德米特里·马尔采夫 费利克斯·穆勒 吉策姆·奥库特 弗里茨·施密德 温 于 2020-08-25 设计创作，主要内容包括：本申请公开了使用机器学习的组合的室内和室外跟踪。本发明涉及一种用于使用跟踪装置的组合的室内和室外跟踪的计算机实现方法,其中,由在待确定的位置处的装置生成无线电信号的指纹。通过将经训练的函数应用于指纹来确定装置的位置,其中,经训练的函数已经使用在已知位置处生成的多个指纹进行了端对端地训练。可以使用环境传感器数据来预测由跟踪装置跟踪的部件的使用期。(The application discloses combined indoor and outdoor tracking using machine learning. The present invention relates to a computer-implemented method for combined indoor and outdoor tracking using a tracking device, wherein a fingerprint of a radio signal is generated by the device at a location to be determined. The location of the device is determined by applying a trained function to the fingerprint, where the trained function has been trained end-to-end using a plurality of fingerprints generated at known locations. Environmental sensor data may be used to predict the life of a component tracked by a tracking device.)

1. A computer-implemented method for determining a position (3) of an apparatus (1), comprising:

obtaining a fingerprint (4) of a radio signal (2) received by the device (1) at a location (3) to be determined;

determining the location (3) of the apparatus (1) by applying a trained function to the fingerprint (4), the trained function having been trained using training fingerprints of radio signals (2) received at a plurality of known locations; and

the determined position (3) is provided.

2. The method according to claim 1, wherein the device (1) further comprises at least one environmental sensor (7), the method further comprising:

obtaining environmental sensor data of the at least one environmental sensor (7);

determining a health condition of the apparatus (1) using the environmental sensor data; and

providing the determined health condition of the device.

3. The method according to claim 2, wherein determining the health of the device (1) comprises:

applying a trained function to the environmental sensor data, the trained function having been trained using environmental sensor training data for a plurality of devices having known health conditions.

4. A method according to claim 2 or 3, wherein the health condition comprises a period of use of a component tracked with the apparatus (1).

5. Method according to one of the preceding claims, wherein applying a trained function to the fingerprint (4) comprises: applying at least one machine learning classification algorithm to the fingerprint (4) by means of a neural network.

6. Method according to one of the preceding claims, wherein the radio signal (2) comprises a plurality of Wi-Fi signals, and wherein the fingerprint (4) comprises a base station ID and/or a signal strength of each of the Wi-Fi signals.

7. The method according to one of the preceding claims, wherein the radio signal (2) comprises: at least two different types of radio signals (2) selected from Wi-Fi signals, Bluetooth signals, GSM signals, and GPS signals.

8. Method according to one of the preceding claims, wherein the fingerprint (4) is a fingerprint of the radio signal (2) and at least one signal of an environmental sensor (7) arranged at the device (1).

9. The method according to one of the preceding claims, further comprising:

obtaining GPS signals from the device (1); and

pre-selecting a location based on the GPS signal that can be determined by the trained function.

10. Method according to one of the preceding claims, wherein the method is performed at several points in time, wherein each determined position (3) of the device (1) is associated and stored with a corresponding time stamp.

11. The method according to one of the preceding claims, wherein the known location is a location inside a building each associated with a unique location name.

12. A computer-implemented method for providing a trained function to determine a location (3) of a device (1), comprising:

-receiving training fingerprints of radio signals (2) received at a plurality of known locations (3);

applying a function to the training fingerprints, wherein for each training fingerprint a location (3) is generated;

determining a difference between the generated location (3) and the known location;

training the function based on the determined differences; and

providing the trained function to determine a position (3) of the device (1).

13. A computer-implemented method for providing a trained function to determine a health condition of a device (1), comprising:

receiving environmental sensor training data for each of a plurality of devices, each device comprising at least one environmental sensor (7) and each device having a known health condition;

applying a function to the environmental sensor training data, wherein for each device, a health condition is generated;

determining a difference between the generated health condition of the device and the known health condition;

training the function based on the determined differences; and

the trained function is provided to determine a health condition of a device.

14. The method of claim 12, wherein receiving a training fingerprint comprises:

moving the device (1) to each of a plurality of positions (3);

scanning the radio signals (2) at each location (3) and, for each location (3), generating a training fingerprint of the radio signals (2); and

a location name is provided for each location (3).

15. A computing device (200), the computing device (200) configured for localization of a field device (1), the computing device (200) comprising a memory (230), an interface (220) and at least one processor (210), the memory (230) containing instructions executable by the at least one processor (210), wherein execution of the instructions causes the computing device (200) to perform the steps of:

obtaining a fingerprint (4) of a radio signal (2) received by the field device (1) at a location (3) to be determined;

determining the location (3) of the field device (1) by applying a trained function to the fingerprint (4), the trained function being trained using training fingerprints of radio signals (2) received at a plurality of known locations; and

the determined position (3) is provided.

16. A computing device (200) for location of a field device, the computing device being configured to perform one of the methods of claims 2 to 14.

17. A tracking system (100), the tracking system (100) comprising at least one field device (1) and at least one computing device (200) according to claim 15 or 16.

Technical Field

Various examples of the invention relate generally to the positioning of devices, and in particular to the tracking of items in supply chain logistics. Various examples relate in particular to indoor and outdoor tracking using a tracking device utilizing a combination of radio signals and Machine Learning (ML) algorithms.

Background

In supply chain logistics, a typical transportation route may include a number of different transportation service providers, where only minor information about the location and condition of the piece of goods is available through the transportation service providers.

Thus, several problems may occur in the conventional repair/spare part logistics. The transparency of the transportation route of the service/spare part stream may be unreliable or completely missing. For example, it may not be possible to see where the cargo is located, nor the condition of the cargo within the cargo. Thus, the status of the package may be unknown, wherein the condition of the spare part, e.g. damage, may not be known, nor the exact location or time of the hand-over of the cargo at the destination.

Furthermore, it may be difficult to find the cargo inside the building. For example, a service technician who is relying on and waiting for a service item in a hospital may not know when the service item is available on site for installation, or may not know in which building of the hospital the cargo item has been received and stored, let alone the exact room. Thus, the technician may require a significant amount of time to determine the exact location of the package inside the building, e.g., a room, which may increase maintenance costs for the hospital.

Furthermore, loss and damage to the goods may occur during transport from the supplier to the hospital. For example, the spare part may arrive at a hospital (i.e., the end customer), where loss or damage has occurred on the way to the hospital, which may not be visible but may result in a reduced life span of the spare part.

Some of the problems have been solved by suppliers and service providers sending information about the location and delivery of goods, which may be, for example, shipping data provided via Electronic Data Interchange (EDI) standards. Unfortunately, however, this information often proves unreliable and opaque. Existing solutions on the market for tracking packages and goods use a wide variety of sensors. Using GPS, outdoor cargo can be tracked and approximate arrival times obtained. Various transport service providers provide shipment tracking on their home pages or via applications. However, reliable solutions for building interiors, such as interior tracking, cannot be provided.

Accordingly, there is a need for advanced tracking techniques that overcome or mitigate at least some of the above-identified limitations and disadvantages.

Disclosure of Invention

Accordingly, the presented concept of a method aims to provide an advanced method and system for determining a position of a device, which overcomes or alleviates at least some of the above identified limitations and disadvantages.

In the following, the solution according to the invention is described with respect to a claimed method for determining a position of a device and with respect to a claimed device and system for determining a position of a device. Features, advantages, or alternative embodiments herein may be assigned to other claimed objects and vice versa. In other words, the claims directed to the apparatus and system may be amended with features that are described or claimed in the context of methods. In this case, the functional features of the method may be embodied by the target unit of the apparatus or system.

It should be understood that the solution according to the invention is described with respect to a method and system for determining a position of a device by applying a trained function and with respect to a method and system for providing a trained function for determining a position of a device. Features, advantages, or alternative embodiments herein may be distributed over other claimed objects and vice versa. In other words, claims directed to methods and systems for providing a trained function to determine a location of a device may be improved by applying the trained function with features described or claimed in the context of the methods and systems, and vice versa. In particular, in the claims relating to applying the trained function, the trained function may be improved with the features described for the method for providing the trained function. Likewise, the input fingerprint may be improved with the features described for the training fingerprint, and vice versa. In particular, the features described for the environmental sensor training data may be exploited to improve the environmental sensor data for determining the health of a component in delivery comprising the apparatus according to the invention, and vice versa. In particular, the trained functions of the method and system for determining the location of a device may be adjusted according to the characteristics of the method and system for providing the trained functions to determine the location of the device. It will be appreciated that the described techniques may be used to determine the location of delivery of a device or a device comprising according to the present invention.

A method for determining a location of a device includes the following steps, which may be a computer-implemented method.

In a first step, a fingerprint of a radio signal received by a device at a location to be determined is obtained. The radio signals may be received by respective antennas of the device. The device may be located at a location that may not be known, wherein the location of the device may need to be determined. The location may be an indoor location, such as inside a building, or may be an outdoor location outside a building. At this location, the device may receive radio signals present at the location, for example, by using a respective operating unit included in the device (e.g., one or more receivers or transceivers or antennas that may be coupled to a respective controller of the device to wirelessly receive the radio signals).

In various examples, the radio signals may include one or more different radio signals, which may be distinguished, for example, by different frequency bands, and may have one or more different characteristics, such as varying signal strength, maximum and/or minimum signal strength, time course or variation of signal strength, distortion occurring in each signal, and interference between radio signals. In other words, analog characteristics and/or digital information may be derived from the radio signal. One or more antennas may be used to receive radio signals.

The device may generate a fingerprint of the radio signal, in other words the fingerprint is a data set. The fingerprint may comprise a subset of the data received by the radio signal. For example, a fingerprint of a radio signal may include analog or digital information, such as one or more characteristics of the radio signal and/or one or more or all of the received radio signal. For example, the signal strength or the change in signal strength over time of one or more radio signals may be included in a fingerprint. Additionally, digitally available information or data of one or more radio signals, such as a base station ID of a radio signal, may be included in the fingerprint.

In general, obtaining data (e.g., fingerprint or sensor data) may include receiving stored data from an internal or external memory or data storage device, and/or receiving data from a sensor or computing device, the data being data that has been measured, or generated and/or processed, wherein any known method of transmitting or receiving data may be implemented.

For example, the radio signal may be a signal transmitted using Radio Frequency (RF), which is, for example and without limitation, the rate of oscillation of an electromagnetic field in a frequency range of about 20kHz to about 300GHz, which may alternatively be referred to as approximately between the upper limit of audio frequencies and the lower limit of infrared frequencies. These are frequencies at which energy from the oscillating current can radiate as radio waves from the conductor into space. Different types of radio signals (e.g., Wi-Fi signals, bluetooth signals, GSM signals, and GPS signals) may specify different upper and lower bounds for a frequency range.

In another step, the location of the device is determined by applying a trained function to the fingerprint, the trained function having been trained using training fingerprints, i.e. training data sets, of radio signals received at a plurality of known locations. In other words, determining the location of the device comprises applying the trained function to the fingerprint of the location, wherein applying the trained function to the fingerprint provides the determined location. Thus, a fingerprint may be an input to a trained function, wherein a location may be an output of the trained function.

Determining the location of the apparatus may include determining a plurality of possible locations of the apparatus, wherein each of the possible locations may be associated with a corresponding probability of the apparatus being at that location.

The trained function for determining the location of the device may have been trained using any method for providing a trained function to determine the location of the device according to the present disclosure.

Typically, the trained functions mimic the cognitive functions of humans in association with the thoughts of other humans. In particular, by training a function, or in other words a machine learning model, based on a training data set, the trained function is able to adapt to new environments and is able to detect and infer patterns. The parameters of the trained function may be adjusted by means of training. For example, supervised training, semi-supervised training, unsupervised training, reinforcement learning, and/or active learning may be used. Further, representation learning (an alternative term is "feature learning") may be used. In particular, the parameters of the trained function may be iteratively adjusted by several iterations of the training method.

For example, the trained function may be an end-to-end trained function trained using multiple training data sets. The training data set may include input data associated with reference output data, for example, a fingerprint of a radio signal at a known location associated with a location name, or an environmental sensor data set associated with a known health condition. Applying the trained function may be performed by a neural network, which may include a plurality of classifier functions.

In various examples, the trained function may include one or more known machine learning classifiers. Without limitation, the trained functions may be based on one or more of a support vector machine, a decision tree and/or a bayesian network, k-means clustering, Q learning, genetic algorithms, and/or association rules, for example. For example, the neural network may be a deep neural network, a convolutional neural network, or a convolutional deep neural network, an antagonistic network, a deep antagonistic network, and/or a generative antagonistic network, or a model-based machine learning network architecture.

In a further step, the determined position is provided. For example, a location name or other unique location identifier, such as a room name or a point of interest (POI) name, may be determined and provided for the location.

The apparatus may also include at least one environmental sensor or a plurality of environmental sensors. The at least one environmental sensor may be comprised in the apparatus or may be arranged at the apparatus.

The environmental sensor may be, for example, but not limited to, a sensor for measuring acceleration such as shock and vibration, a temperature sensor, an air pressure sensor, an ambient light sensor, a radioactive radiation sensor, a UV exposure sensor, a humidity sensor, an optical sensor, or any other sensor as known in the art.

The method may further include obtaining environmental sensor data for at least one environmental sensor. In other words, the environmental sensor may perform periodic or continuous measurements of environmental conditions of the environment surrounding the device (which may also be referred to as the ambient environment). The measurement signals of the environmental sensors can be read and stored at time intervals in order to provide a time line or sequence of measurement data. The measurement data may be sent to the back-end together with the fingerprint or may be sent to the back-end independently of the fingerprint.

Based on environmental sensor data of the at least one environmental sensor, a health condition of the device may be determined. It will be appreciated that the described techniques may be used to determine the health of an apparatus or a delivery including the apparatus or a component tracked by the apparatus, wherein the component may be included in the delivery, i.e. the health may apply to the apparatus and/or equally to the delivery/component tracked with the apparatus.

In other words, using the environmental sensor data, it may be determined whether one or more of a plurality of predetermined health states or health conditions of the delivery and/or components in the delivery occurred. Further, determining the health condition may include determining a plurality of possible health conditions that are each associated with a corresponding probability that the component is in such a health condition. Health condition may refer to a quality condition of a component, which may be defined as, for example, but not limited to, undamaged, lightly worn, damaged, or unusable. The determined health of the device may be provided to, for example, a backend or transport service provider.

Determining the health condition may include applying a trained function to the environmental sensor data, the trained function having been trained using the environmental sensor training data of the plurality of devices and the known health condition of the device. In various examples, the trained function used to determine the health of the device may be an end-to-end trained function that has been trained using a training data set. The training data set may include known health conditions of the apparatus and corresponding environmental sensor training data, which may be sensor data generated or measured by at least one environmental sensor of the apparatus. The at least one environmental sensor may be attached to or arranged at the device, or may be comprised in the device.

For example, a training data set may have been created over a period of time for a plurality of components that are each shipped and tracked with a respective tracking device. Environmental sensor data may have been collected during the transportation or transport of each of the devices. Furthermore, after the component is transported, the health condition may have been determined by inspection or testing of the component. In various examples, a component may have been running for a period of time, where pre-broken lifetime data for the lifetime (lifetime) of the component or the lifetime of the component after a component failure has been determined, where the health condition may include the lifetime of the component. The training data set for providing the trained function to determine the health of the component may include environmental sensor data collected during transportation and a corresponding health or age of the component.

By using environmental sensor data, the health of the component after transport may be predicted using tracking techniques according to the present disclosure. For example, environmental sensor data may be used to alert a field technician receiving a shipment that he should scrutinize the received shipment.

Replacement delivery for new spare parts may be initiated based on notification of a damage event for a cargo part before a defective spare part that has been damaged during transport reaches its destination. This can minimize the waiting time until the arrival of a functional spare part. The liability problem can be clearly solved. This ensures traceability and documentation; the use may enable traceability for outdoor or indoor locations. A combination of two measurement systems, GPS and Wi-Fi fingerprinting (fingerprint) may be used for data acquisition.

In addition to tracking of ambient environmental conditions around the device, which may enable health condition tracking that can be provided for the device being transported, a life expectancy prognosis may also be provided.

Environmental sensor data may be used in a machine learning model to predict the life of materials or components included in a shipment. Using machine learning techniques, i.e., trained functions trained using measured cargo conditions and actual life of the device, life may be more reliably predicted and effects on life may be identified based on a number of ambient conditions of the device during transit. Over time, the accuracy of such lifetime predictions may be increased using previous shipping experience of the device. To improve future life expectancy of another component, the received environmental sensor data may be stored in a database as input training data for training the trained function. When at a subsequent point in time, the component develops a defect and thus the lifetime of the component is known, the environmental sensor data can be associated with the component and the component lifetime, and can be used to train the machine learning model to predict the lifetime of other components.

In the following, as a separate method, a method for providing a trained function for determining a position of a device is provided, which may be performed independently of a method for determining a position of a device using a trained function.

A method for providing a trained function to determine a location of a device includes the following steps, which may be a computer-implemented method.

In a first step, training fingerprints of radio signals received at a plurality of known locations are received. At each known location, in a similar manner as described for the method for determining the location of the apparatus, a radio signal present at that location may be received and a fingerprint of the radio signal may be generated.

In another step, for each training fingerprint, a corresponding location is generated. Generating a location for a training fingerprint includes: a function or trainable function is applied to the training fingerprint to generate a location.

In another step, a difference between the generated position of the device and the known position of the device is determined. For each apparatus, a difference between the generated position and the known position may be generated, and based thereon, the difference may be determined as an overall difference.

In a further step, a function is trained based on or in other words using the determined differences. Training the function may include adjusting parameters in the function in a manner that minimizes the difference.

In another step, a trained function for determining the position of the device is provided.

For example, receiving the training fingerprint may include: the apparatus is moved to each of a plurality of known locations, wherein at each known location the apparatus scans for radio signals, i.e. receives radio signals present at that location, and at each location generates a fingerprint of the radio signals. Further, the apparatus may provide a fingerprint for each known location as a training data set for the machine learning function. For example, a location name may be provided for each location, where the fingerprint may be associated with a known location and/or location name in order to form a training fingerprint for training a machine learning model to determine the location of the apparatus.

A computer-implemented method for providing a trained function to determine a health condition of a device includes the following steps.

In a first step, environmental sensor training data for each of a plurality of devices is received. Each device includes at least one environmental sensor and a component that is tracked with the device in a known health condition after exposure to an environmental condition measured by the sensor during transport.

In another step, a function is applied to the environmental sensor training data, wherein for each device, a health condition is generated. For example, as in other methods for providing a trained function according to the present disclosure, the trained function may be an end-to-end trainable function. The health of each device is generated as an output by applying a trainable function to environmental sensor training data for the plurality of devices.

In another step, a difference between the generated health condition and a known health condition of the component is determined. In particular, for each of the apparatuses, a difference between the generated health condition and a known health condition may be determined in pairs, wherein the total difference is determined based on the pair-wise determined differences.

In another step, a trainable function is trained based on the determined differences.

In another step, a trained function for determining the health of the component is provided.

The technique according to the present disclosure may have the following advantages. Without the need to build an infrastructure system, building an infrastructure system would involve significant cost and high expense, particularly for the building owner.

This technique may provide greater location accuracy because the location and condition of the shipment may be specified more accurately than may be specified using conventional methods (e.g., GPS). Here, artificial intelligence using machine-learned Wi-Fi fingerprinting techniques according to the present disclosure provides a significant improvement in indoor positioning by being able to effectively process any information from incoming radio signals (including, for example, weaker radio signals or even interference and disturbances).

Accurate indoor positioning can be performed using an AI engine that processes and filters radio signal data and environmental sensor data, using both analog and digital information of the radio signal. The determined location may be specified using the existing room number and name of the building. Furthermore, damage to the cargo item can be identified more quickly.

The techniques according to the present disclosure may be applied in other areas than service logistics and may use additional interfaces to address a wide range of issues associated with logistics, such as warehouse logistics, or more generally, material flow at an end customer (e.g., a hospital or other institution).

Not only can delivery delays be detected, but optimized delivery routes can be provided in the future and possible problems during the life of the component can be predicted in advance. With the aid of the optimization of the transport route, costs can be saved due to the perfect time management. In addition, this may lead to increased efficiency in view of the situation of technicians in a hospital, for example, as knowledge from accurate location tracking may be used to improve resource planning. In addition, all of the listed advantages enable high satisfaction of the end customer, which may be beneficial to the relationship with the contracting partners.

A computing device is configured for determining a location of a field device and includes a memory, an interface, and at least one processing unit, the memory containing instructions executable by the at least one processing unit, wherein execution of the instructions causes the computing device to perform the following steps.

In a first step, a fingerprint of a radio signal received by a field device at a location to be determined is obtained by a computing device. In another step, the computing device determines a location of the field device by applying a trained function to the fingerprint, wherein the trained function is trained using training fingerprints of radio signals received at a plurality of known locations. In another step, the computing device provides the determined location.

The computing device may also be configured to perform the steps of any method or any combination of methods in accordance with the present disclosure.

In accordance with the present disclosure, a system or tracking system is configured to perform any method of determining a location of a field device using a trained function, or providing a trained function to determine a location of a field device. The system includes at least one computing device and at least one field device according to the present disclosure.

A computer program product and a computer-readable storage medium include program code to be executed by at least one processor of a computing device. Wherein execution of the program code in the at least one processor causes the computing device to perform one of the following methods in accordance with the present disclosure: methods for providing a trained function to determine a location of a device or methods for determining a location of a device using a trained function.

For such a computing device, tracking system, computer program product and computer readable storage medium for determining a position of a device using a trained function and for providing the trained function for determining the position of the device, a technical effect may be achieved, which corresponds to the technical effect described for the method of: methods for determining a location of a device using a trained function and for providing a trained function to determine a location of a device according to the present disclosure.

Although specific features described in the foregoing summary and the following detailed description are described in connection with specific examples, it should be understood that the features may be used not only in the respective combinations but also individually, and that features from different examples may be combined with each other and related to each other, unless specifically stated otherwise.

Accordingly, the above summary is intended only to give a brief summary of some embodiments and some features of implementations and is not to be construed as limiting. Other embodiments may include other features in addition to those described above.

Drawings

In the following, the concept according to exemplary embodiments of the invention will be explained in more detail with reference to the following drawings:

fig. 1 schematically shows a typical transport route for spare parts in a service/spare part logistics;

FIG. 2 schematically illustrates a tracking system for determining a location of a field device using a trained function, in accordance with an embodiment of the present invention;

FIG. 3 schematically illustrates a combined outdoor and indoor tracking system with health tracking, according to an embodiment of the invention;

FIG. 4 shows a flow diagram of a method for determining a location of a device using a trained function, according to an embodiment of the invention;

FIG. 5 illustrates a flow diagram of a method for providing a trained function to determine a location of a device, according to an embodiment of the invention;

FIG. 6 illustrates a flow diagram of a method for providing a trained function to determine a health of a component, according to an embodiment of the invention; and

FIG. 7 illustrates a schematic diagram of a computing device configured to determine a location of a field device using a trained function, according to an embodiment of the invention.

Detailed Description

The above and other elements, features, steps and concepts of the present disclosure will become more apparent from the following detailed description of exemplary embodiments according to the present invention, which is to be read in connection with the accompanying drawings.

Some examples of the disclosure generally provide for a plurality of circuits, data storage devices, connections, or electrical devices such as, for example, processors. All references to these entities or other electrical devices or the functionality provided by each are not intended to be limited to encompassing only what is shown and described herein. While particular labels may be assigned to the various circuits or other electrical devices disclosed, such labels are not intended to limit the operating range of the circuits and other electrical devices. Such circuits and other electrical devices may be combined and/or separated from one another in any manner based on the particular type of electrical implementation desired. It is recognized that any circuit or other electrical device disclosed herein may include any number of microcontrollers, Graphics Processor Units (GPUs), integrated circuits, memory devices (e.g., flash memory, Random Access Memory (RAM), Read Only Memory (ROM), Electrically Programmable Read Only Memory (EPROM), Electrically Erasable Programmable Read Only Memory (EEPROM), or other suitable variations thereof), and software that cooperate with one another to perform the operations disclosed herein. Additionally, any one or more of the electrical devices can be configured to execute program code embodied in a non-transitory computer readable medium that is programmed to perform any number of the disclosed functions.

It will be understood that the following description of embodiments is not to be taken in a limiting sense. The scope of the invention is not intended to be limited by the embodiments described below or by the drawings, which are to be considered illustrative only.

The figures are to be regarded as schematic representations and elements shown in the figures are not necessarily shown to scale. Rather, the various elements are represented such that their function and general purpose are apparent to those skilled in the art. Any connection or communication or coupling between functional blocks, devices, components or other physical units or functional units shown in the figures or described herein may also be achieved through indirect connections or couplings. Communication between devices may also be established through a wireless connection. Functional blocks may be implemented in hardware, firmware, software, or a combination thereof.

In the following, techniques are described that may facilitate monitoring and positioning of cargo, particularly inside buildings, in the area of supply chain logistics.

Fig. 1 schematically shows a typical transport route for spare parts in a service/spare part stream.

Referring to fig. 1, the transportation of replacement CT tubes to a hospital is depicted. It will be appreciated that similar principles and problems may be applied to, for example, any material flow, such as transporting parts to or in an industrial production line.

In step L1, a CT tube in the hospital develops a defect. In step L2, spare, i.e. replacement, CT tubes are packaged at the supplier for transport to the hospital. In step L3, the spare CT tubes are transported to the airport. In step L4, the spare CT tubes are transported by airplane. In step L5, the spare CT tube is delivered to the distributor, who will deliver and hand over the spare CT tube to the hospital. In step L6, a replacement CT tube is installed at the hospital by a technician of the CT tube supplier replacing the defective CT tube. In step L7, the defective CT tube may be returned to the supplier in the opposite direction of the logistics path.

As indicated by the dashed box in fig. 1, after the replacement CT tube is packaged at the supplier and handed to the first transport service provider, and until the spare tube is received by the technician of the supplier on site in the hospital, the exact information is not known, where the spare CT tube is located, and under what circumstances the spare CT tube is. In this regard, the logistics path between steps L2 and L6 may be considered as a black box, wherein only little information is available through the transport service provider.

Several problems may be associated with the supply chain logistics described above. The transparency of the transportation route with respect to the logistics of service/spare parts may be unreliable or completely missing. For example, it may not be possible to see where the cargo is located, nor the condition of the cargo within the cargo. Thus, the status of the package may be unknown, wherein the condition of the spare part, e.g. damage, may not be known, nor the exact location of the cargo part.

Furthermore, it is difficult to find cargo inside buildings. Technicians that rely on and wait for the shipment may not know when the device is available for installation on site, or may not know in which exact room of the hospital or even the building the shipment has been received and stored. Thus, the technician may require a significant amount of time to determine the exact location of the package inside the building, e.g., a room, which increases maintenance costs for the hospital.

Such a typical supply chain may include a plurality, in some examples up to twenty-six different logistics service providers, where, in particular, handing over at a hospital and storing in a hospital may often not be clearly defined. Thus, the necessary spare parts may be located at an unknown location inside the building.

Furthermore, loss and damage to the cargo may occur during transport from the supplier to the hospital. For example, a defective spare part may arrive at a hospital (i.e., an end customer), where it has been lost or damaged on its way to the hospital.

This may require subsequent clarification, e.g. the warranty, i.e. responsibility, of the responsible party may have to be determined. Further, the alternate delivery is only made after knowing that the package has defectively reached its destination, rather than at the earliest possible time that damage/loss occurs. As a result, extended downtime may occur, thereby reducing customer satisfaction. For example, a hospital or physician may increase machine costs and increase technician working time due to downtime of the medical device.

Heretofore, some problems have been addressed by having suppliers and service providers provide information about the location and delivery of items (e.g., providing shipping data via EDI). Unfortunately, in the past, these illustrations have proven unreliable and opaque.

Existing solutions on the market for tracking packages and goods use a wide variety of sensors. Using GPS, outdoor cargo can be tracked and approximate arrival times obtained. Various transport service providers provide shipment tracking on their home pages or via applications. However, such solutions for building interiors, such as interior tracking, cannot be provided.

Fig. 2 schematically shows a tracking system 100 for determining the position of the apparatus 1 using a trained function according to an embodiment of the invention.

As depicted in fig. 2, the tracking system comprises a field device 1, which field device 1 may be referred to as the front end of the tracking system and is located at a location 3 to be determined. At location 3, field device 1 is exposed to various radio signals, in this example including one or more GSM signals 2 from one or more cellular network base stations or radio signals typically used for mobile communications in a cellular network, one or more bluetooth signals 2, and one or more Wi-Fi signals from one or more Wi-Fi base stations. The device 1 communicates with a computing device 200, which computing device 200 may also be referred to as the back end of the tracking system 100.

Referring to fig. 2, the machine learning function according to the present invention will be explained in more detail. By combining different radio signals 2 (e.g. Wi-Fi, bluetooth and GPS/GSM), the tracking device 1 can be used to provide a room-based location of a piece of goods. In the context of the present disclosure, a tracker may equally be referred to as a tracking device, a field device, or a device. Wherein the tracker may for example be attached to or comprised in a piece of goods, such as a spare part or a package.

The radio signal at the specific location 3 may include the following information: e.g. signal strength and base station ID of a Wi-Fi signal, which is received by the device 1 and further transmitted from the device 1 to the computing device 200 providing the AI engine as the fingerprint 4 of the radio signal 2 at the location 3. The AI engine provided by the computing device 200 at the back-end implements an analysis function using machine learning for the fingerprint 4 (i.e., applies a trained function to the fingerprint 4), and can train the function using a training fingerprint having an associated known location. By applying the trained function to the fingerprint 4, the position 3 of the field device 1 is determined.

The fingerprint 4 sent to the AI engine includes a list of found radio signals 2, the corresponding signal strength of each signal and optionally the GPS coordinates of the device at the current location 3.

Using GPS coordinates, the tracking device 1 can pre-select the possible locations of the package inside the building. A known or determined location may be represented by a location name 5, for example by an address and/or a room number or name inside a building. The determined location name 5 is sent to the field device 1.

In the training process for training the function, the site must be scanned in advance, and thus a training fingerprint must be set for a known position. For example, these fingerprints may be scanned and stored by a technician using a mobile application, and may then be provided to a backend along with the name of the location, in order to provide a training data set for training a function to provide the device location.

In this manner, a virtual infrastructure is created using an AI engine that can then provide accurate information about the location of the package. Wherein the AI engine uses a fingerprint 4 of the radio signal 2 comprising the signal strength and the base station ID and applies a trained function to the fingerprint 4 in order to generate at least one location. The AI engine then provides the probability of the most recent fingerprint (i.e., the location associated with the known fingerprint) and in this way discloses the location of the package.

The pre-selection can be made using GPS coordinates, which helps to ensure scalability of our solution. For example, when identifying that a package is being transported to a city in germany, using outdoor tracking with GPS, the AI engine will only use the relevant subset of fingerprints located in the corresponding city in germany. The AI engine uses at least one machine learning function or classifier as known in the art, such as, for example, SVMs and/or neural networks. In various examples, in a hierarchical network architecture, the AI engine uses a variety of machine learning functions, e.g., seven or more machine learning functions.

Hereinafter, the machine learning function will be described in further detail.

The technician is located at a point of interest (POI) inside the building and uses a button in his application called "add POI (add POI)". After pressing, the radio signals in the current location are scanned and sent to the backend structure, which may include an AI engine. In addition, the technician enters the appropriate and detailed room name associated with the current location. POIs are named by this name to achieve the exact location later. In other words, a training fingerprint is generated, the training fingerprint comprising a location name and a fingerprint of the radio signal at the location. The training fingerprint is stored at the backend along with a plurality of other training fingerprints that are associated with different POIs and that may initiate a learning process for AI, i.e., the training fingerprint may be used to generate or train a machine learning model.

Using a trained function trained with fingerprints of known locations, the package can then be located. Wherein the tracker operates as described below. The tracker scans for radio signals of its environment, queries the data by the back end of the AI engine to find a location, and then displays the location to the technician in the application.

To reduce the amount of data to be compared, fingerprints may be grouped into families. The families may be grouped and/or selected based on their GPS coordinates.

Fig. 3 schematically illustrates a combined outdoor and indoor tracking system 100 with health tracking according to an embodiment of the present invention.

The shipment comprises a package with a tracking device 1, which is transported and delivered to a hospital. The tracking device 1 includes a plurality of environment sensors 7, and the environment sensors 7 include a temperature sensor 7, a humidity sensor 7, an acceleration sensor 7, and a pressure sensor 7. The tracking device 1 further comprises at least one receiver for receiving the radio signal 2.

The tracking device may be used with shipping or delivery to track repair or spare parts during transportation.

Outdoor tracking may be provided via GPS, with transmission via GSS. During outdoor transport, the tracking device 1 generates a fingerprint from the radio signal received at the location. Furthermore, the tracking device 1 collects measurement data of the environmental sensor 7.

Indoor tracking may be provided by setting points of interest (POIs) using so-called Wi-Fi fingerprinting via Wi-Fi based end-to-end trained machine learning functions. When the item of cargo has been delivered to the hospital and transported and stored inside the hospital building, the tracking device collects the fingerprint of the radio signal, including the Wi-Fi signal 2. In this way, artificial intelligence is used to create a virtual infrastructure inside a building.

Using the environmental sensor 7, shocks and vibrations (i.e. acceleration sensor limit violations, temperature limit violations, air pressure violations, etc.) are detected and sent to the application, which in turn can send data to the back-end.

The tracking device 1 sends the fingerprint and collected environmental sensor data to a database or cloud data storage 8 at time intervals. The computing device 200 uses the data stored in the cloud storage 8 to determine the location of the tracking device 1 for each fingerprint. Further, the health of the package is determined based on the environmental sensor 7 data.

As depicted in fig. 3, and described in further detail, techniques according to the present disclosure utilize various sensors and machine learning to provide not only location data for outdoor use, but also accurate location data for indoor use.

Using the determined position of the tracking device 1, which is displayed to the technician 9 as a time line of positions for example, the technician may be able to easily find the piece of goods inside the hospital building. Wherein the application, which may be an application of the mobile electrical device, may visualize the positioning and location to a user, e.g. a technician 9. Further, the application may provide various types of configuration options for the user.

By storing the events that occur on the shipping route, for example, on the memory card of the tracker, the liability issue can be clearly resolved. This ensures traceability and documentation; the use may enable traceability for outdoor or indoor locations. A combination of the two measurement systems (i.e., GPS and Wi-Fi fingerprinting) may be used for data acquisition.

Replacement delivery of new spare parts may be initiated based on notification of a damage event for a cargo part before the defective spare part that has been damaged during transport reaches its destination. This can minimize the waiting time until the arrival of a functional spare part.

The technique according to the present disclosure may have the following advantages. Without the need to build an infrastructure system, it would involve significant costs and high expenditures, especially for the building owner.

This technique may provide greater location accuracy because the location and condition of the shipment may be specified more accurately than may be specified using conventional methods (e.g., GPS). Here, artificial intelligence using machine-learned Wi-Fi fingerprinting techniques according to the present disclosure provides significant improvements for indoor positioning.

The AI engine, which processes and filters the obtained data, can be used to perform accurate indoor positioning using both analog and digital information of the radio signal 2. The determined location may be specified using the existing room number and name of the building. Furthermore, damage to the cargo item can be identified more quickly and accurately using the data of the environmental sensor 7.

In various examples, fingerprinting may refer to fingerprints that generate radio signals and/or environmental sensor signals from Wi-Fi, bluetooth, and GSM receivers.

The fingerprint may be sent through a communication structure of the cloud storage that may be selected independently of the transport service provider, resulting in full data rights for the vendor and/or the end customer.

The techniques according to the present disclosure may be applied in other areas than service logistics and may use additional interfaces to address a wide range of issues associated with material flow, for example, in warehouse logistics, or more generally at an end customer (e.g., a hospital or other institution).

Not only can delivery delays be detected, but the storage of fingerprints and the use of artificial intelligence, particularly machine learning, can provide optimized delivery routes in the future and predict potential problems ahead of time during the lifetime of the component. With the aid of the optimization of the transport route, costs can be saved due to the perfect time management. In addition, this may lead to increased efficiency in view of the situation of technicians in a hospital, for example, as knowledge from accurate location tracking may be used to improve resource planning. In addition, all of the listed advantages enable high satisfaction of the end customer, which may be beneficial to the relationship with the contracting partners.

For example, a neural network and support vector machine may be deployed to process fingerprints. For example, a sequence definition algorithm such as LSTM may be deployed for analysis of sensor data.

The tracking system 100 according to the present disclosure may comprise a front end with one or more tracking devices 1 transmitting fingerprints 4 to a back end comprising at least one computing device 200 to perform a machine learning function. In addition, web applications and/or desktop applications can be provided to display and provide information and functionality, i.e., application views, to a technician.

Hereinafter, an application view according to the present invention will be explained in further detail.

The front end portion for outdoor/indoor positioning may display an administrator view. The data is sent to and received from the AI engine through the tracker with the database and the location data is represented for outdoor positioning.

An overview (overview) map may be displayed having a pointer and indicating the current position of the tracker and the accuracy of the current position by a circular pointer around the position. The tracker updates whenever it sends updated information about its location.

A location history map may be displayed containing all locations where the tracker has been located and sent data, which may be presented as an indicator with a circle of precision (accuracies) around the perimeter connected with a line indicating movement from one point to another. When the mouse is suspended, the coordinates and various sensor information such as temperature, humidity and the like are displayed.

In the indoor positioning view, multiple columns may be displayed, including POIs (points of interest, which may include the name of a location inside any facility), probability percentages (of the tracker being at that exact location), and timestamps.

The front end portion for outdoor/indoor positioning may display a technician view, as will be described below.

The tracker sends data to the AI engine along with the database and receives location data from the AI engine and the location data is represented for outdoor positioning.

An overview map may be displayed with an indicator indicating the current position of the tracker and a surrounding circle of accuracy. Updates are displayed whenever a tracker sends update information about its location.

In the indoor positioning view, multiple columns may be displayed, including POI (point of interest, which will be the name of the location inside any facility), probability percentage (of the tracker being at that exact location), and timestamp.

The "add poi (add poi)" function may be presented only in mobile applications. The technician writes the name of the indoor location and the application starts scanning the indoor location for radio signals, creating a fingerprint of the location. Then, if the tracker is to be within that location, the probability of being within that POI will be shown in the indoor positioning view.

Referring back to fig. 3, a combination of outdoor/indoor tracking systems and Wi-Fi fingerprinting technology is provided as a specific method for indoor tracking. Wherein an outdoor tracking or an outdoor positioning can be provided with an accurate positioning determined by GPS and transmitted via GSM, wherein the data transmission can be set separately. Furthermore, indoor positioning is provided, wherein Wi-Fi fingerprinting is used to find a second location in a building without the need to build expensive and complex infrastructure. Indoor tracking via Wi-Fi fingerprinting is achieved by a machine learning system.

Further, health or condition tracking may be provided, wherein real-time tracking of events may be performed using multiple environmental sensors. The measurement data of the environmental sensors can be analyzed and further processed using the AI system. The measurement data may be included in a training fingerprint and a fingerprint for a machine learning function. In particular, certain events, i.e., fault or damage events, may be trained by the machine learning function and predicted by the trained machine learning function. For example, a trained machine learning function may be used to provide a prognosis of the life of a spare part. Thus, early detection of possible damage may be provided. Full ownership of data (e.g., information about products, transportation, logistics network/supply chain, and/or packaging) that can be used in a machine learning system can be obtained, where such data can be used to train machine learning functions for the above predictions and prognoses. In this manner, the techniques according to the present invention may combine three parts, outdoor, indoor, and health/condition tracking.

Fig. 4 shows a flow chart of a method for determining the position of the apparatus 1 using a trained function according to an embodiment of the invention.

The computer-implemented method for determining a location of a device begins in step S10. In step S20, a fingerprint of a radio signal received by a device at a location to be determined is obtained. In step S30, the location of the device is determined by applying a trained function to the fingerprint, which is trained using training fingerprints of radio signals received at a plurality of known locations. In step S40, the determined position is provided. The method ends in step S50.

Fig. 5 shows a flow chart of a method for providing a trained function for determining the position 3 of the apparatus 1 according to an embodiment of the invention.

The computer-implemented method for providing a trained function to determine the location of a device begins in step T10. In step T20, training fingerprints are received, wherein each training fingerprint is a fingerprint of a radio signal received at a known location. The training fingerprint may be obtained along with or in association with a known location (e.g., location name). In step T30, a function is applied to the training fingerprints, which may be an end-to-end trainable function, wherein for each training fingerprint a corresponding location is generated and output. In step T40, a difference between the generated position and the known position is determined. Generating the difference may include: a difference generated based on a distance between the generated location of the fingerprint and the corresponding known location. In step T50, a function is trained based on the determined difference. In step T60, a trained function for determining the position of the device is provided. The method ends in step T70.

Steps T30, T40 and T50 relate to a learning phase, or in other words a training phase of a trained function, which can be iterated in order to train the function, or when a new training data set is available. In an application phase, the trained function trained in the learning phase is applied to the fingerprint of the unknown location to be determined.

FIG. 6 illustrates a flow diagram of a method for providing a trained function to determine a health of a component, according to an embodiment of the invention.

The method for providing a trained function to determine the health of a component begins in step U10. In step U20, environmental sensor training data is received. The environmental sensor training data includes environmental sensor data for each of the plurality of devices associated with a known health of the component tracked with the device. Each device may include at least one environmental sensor. In step U30, a trainable function is applied to the environmental sensor training data, wherein for each device, a health condition is generated. In step U40, a difference between the generated health condition and a known health condition of the component is determined. The difference may be determined, wherein for each device, the generated health condition is compared to a known health condition of the respective component. In step U50, a trainable function is trained based on the determined differences. In step U60, a trained function for determining the health of the component is provided. The method ends in step U70.

Fig. 7 shows a schematic diagram of a computing device 200 for determining the position of a field device 1 using a trained function according to an embodiment of the invention.

The device or computing device 200 is configured for localization of a field device and/or for providing a trained function to localize the field device, wherein the computing device 200 comprises a memory 230, an interface 220 and at least one processing unit 210. Wherein the memory 230 contains instructions executable by the at least one processing unit 210, wherein execution of the instructions causes the computing device 200 to perform steps according to the method described with respect to fig. 4 and 5.

From the above, some general conclusions can be drawn:

techniques according to the present invention may provide combined indoor and outdoor tracking of cargo. Indoor tracking may be provided using machine learning through advanced Wi-Fi fingerprinting or sniffing (sniff) techniques with improved accuracy and capable of more accurately distinguishing between multiple locations than conventional indoor tracking systems.

For example, a trained function for determining the location of the device may be applied and trained with a data set comprising at least one or all of the environmental sensor data and a fingerprint of the radio signal. For example, a trained function for providing a health condition of a component may be applied and trained with a data set including radio signal fingerprints generated by the apparatus during transportation. For example, the fingerprint may be a fingerprint of a radio signal received by the apparatus and at least one environmental sensor signal of the apparatus.

Additional information in the data set for the machine learning model may also improve the accuracy of the output of the machine learning model.

Referring back to the localization technique, applying the trained function to the fingerprint may include applying at least one classification algorithm to the fingerprint, which may be a machine learning classifier. The application of the trained function may be performed using a neural network. For example, the trained function may be a machine-learned classifier applied to data sets in multiple layers in a neural network.

In various examples, the radio signals may include a plurality of different Wi-Fi signals, i.e., signals of various Wi-Fi base stations, wherein the fingerprint includes a signal strength and/or a base station ID of each of the Wi-Fi base stations.

The radio signals may comprise at least two different types of radio signals, wherein the type of radio signals may refer to Wi-Fi type radio signals according to, for example, IEEE 802.11x industry standards, or bluetooth type radio signals or global system for mobile communications (GSM) type radio signals or Global Positioning System (GPS) type signals.

The method for determining the position of the device may further comprise: obtaining a GPS signal from the device, and pre-selecting a location based on the GPS signal that can be determined by the trained function. When the device may be determined to be located near a destination in a certain country or city or building based on GPS signals, the machine learning algorithm for determining the location may be configured to output only locations that match or are a certain range from the location determined based on GPS signals. For example, only locations or rooms in a particular building may be pre-selected for output.

The method for determining the position of the device may be performed for the device a plurality of times during transportation. For each determination, the determined location of the apparatus may be stored in the database in association with a corresponding timestamp. Using a timeline of locations traversed by the apparatus, the apparatus may be found by a person, for example by starting a search from the last determined location, even though the last location of the apparatus may not be reliably determined.

In particular, the location and the known location in the training dataset may be indoor locations, e.g. inside a building, each associated with a unique location name. The method according to the invention may advantageously provide indoor tracking of the interior of a building, wherein the accuracy of the location is increased compared to conventional Wi-Fi fingerprinting techniques.

In other words, the trained function may be trained using multiple points of interest or significant locations in the building. When the location of the device cannot be determined with high probability in the application phase of the trained function, a timeline of earlier locations may be stored for the device, and a technician may use the timeline to track the path of the shipment through the building and find the shipment. For example, a trained function might determine multiple possible locations, and then interpolate between the locations to determine a single location of the device.

Using the disclosed machine learning techniques, the field device itself can determine its location.

For example, the radio signal may be a strong signal including digital information such as a sender ID, or the radio signal may be any other signal that does not include digital information and that cannot be conventionally used for positioning, such as a jamming signal, a weak or single signal, an intermittent signal, a pulse-like or interrupted signal, a bluetooth signal, and even the environmental sensor signal may be included in fingerprints and training fingerprints, which improves the accuracy of positioning by machine learning techniques.

The fingerprint may correspond to a particular location because radio signals and/or environmental sensor signals are received and/or measured at the particular location. For example, it is possible that at least one non-radio signal, e.g. a sensor signal of an environmental sensor, may be included in the fingerprint and used by a trained function to improve the positioning.

The movable device (in other words, the field device or the tracking device) may be referred to as the front end or front end portion of the tracking system. The computing device and/or database may be referred to as a back end or back end portion of the tracking system, which is remote from the tracking device and wirelessly receives the input fingerprint from the tracking device and determines and/or stores the location and/or health of the item including the tracking device.

The radio signals in the fingerprint may be received by a device at the location to be determined. In other words, the device is in a position where the position is unknown, and it is to be determined at which position the device is located.

The trained function may be trained using a plurality of training fingerprints from known locations, where each training fingerprint may include data measured or received at a single location of the plurality of known locations, and may thus be associated with a known location. Thus, as is known in the art, the trained function may be an end-to-end trained function. Where a fingerprint may be referred to as an input data set, and a location may be referred to as an output data set of a trained function or machine learning model.

Determining the location of the apparatus may comprise determining a plurality of possible locations of the apparatus and determining, for each location, a probability that the apparatus is at that location, such that each location is associated with a probability that the apparatus will be at the respective location. The method may further include displaying a list of possible locations and corresponding probabilities to a user.

The fingerprint or fingerprints may comprise a subset of information obtainable both in an analog and in a digital manner from the radio signal, such as the presence and signal strength of interfering signals, the frequency or delay or channel of each signal.

The indoor location name may be, for example, a room name or room number of a building.

In summary, a computer-implemented method for determining a position of a tracking device is provided in which a machine learning function may be used to more accurately predict the position of the tracking device inside a building. For example, indoor tracking methods and systems may quickly guide a technician to a delivered shipment inside a building. Furthermore, the machine learning function may advantageously be used to alert technicians of quality risks and predict the life of transportation spares.

Thus, techniques according to the present disclosure may enable combining the use of outdoor and indoor location data compiled through Wi-Fi fingerprinting and AI sensor data extracted using intelligent tracking devices.

Although the invention has been shown and described with respect to certain preferred embodiments, equivalent alterations and modifications will occur to others skilled in the art upon the reading and understanding of this specification. The present invention includes all such equivalent variations and modifications, and is limited only by the scope of the following claims.