阅读说明：本技术 监测机器人机械状况的装置和方法 (Device and method for monitoring mechanical condition of robot ) 是由 A·马亚里 N·森 T·洛克 高文智 戴丹 S·扎赫拉伊 于 2019-08-21 设计创作，主要内容包括：本发明涉及监测机器人机械状况的装置和方法。机器人的机械状况监测可以用于检测机器人的意外故障。处理取自机器人操作的数据,并且将该数据与健康基线进行比较。在机器人操作的监测阶段期间提取的特征与在训练阶段期间提取的特征对准,其中健康基线通过将上述两者投影到共同子空间上来建立。可以包括距离评估(诸如L2范数)的分类器被用在公共子空间内以评估机器人的状况。距离评估离标准的偏移指示故障或潜在故障。(The invention relates to a device and a method for monitoring the condition of a robot machine. Mechanical condition monitoring of the robot may be used to detect unexpected faults of the robot. Data taken from the robot operation is processed and compared to a healthy baseline. Features extracted during the monitoring phase of the robot operation are aligned with features extracted during the training phase, wherein a healthy baseline is established by projecting both onto a common subspace. Classifiers, which can include distance evaluations (such as the L2 norm), are used within a common subspace to evaluate the condition of the robot. An offset of the distance assessment from the criteria indicates a fault or potential fault.)

1. An apparatus, comprising:

a robot having components configured to undergo relative mechanical movement during operation of the robot;

a sensor coupled with the robot and configured to detect an operating condition of the component;

a diagnostic device configured to receive measurements from the sensors and including a memory having training data, the diagnostic device configured to distinguish between changes in health of the robot and changes in operation of the robot, and having computer-based instructions, the diagnostic device configured to:

calculating a principal component analysis of the measurements from the sensors to provide operational data;

determining a common subspace between the training data and the operational data; and

in the common subspace, utilizing a classifier trained on the training data and applied to the operational data.

2. The apparatus of claim 1, wherein the diagnostic device comprises a computer processing device and a computer memory, wherein the memory includes the computer-based instructions which, when executed by the computer processing device, operate to provide an output signal indicative of a distance between the training data and the operational data in the common subspace, the output signal having greater sensitivity to changes in the health of the robot and less sensitivity to changes in the operation of the robot.

3. The apparatus of claim 1, wherein the diagnostic device is further configured to determine the common subspace via unsupervised domain adaptation.

4. The apparatus of claim 3, wherein the unsupervised domain adaptation is performed using manifold alignment.

5. The apparatus of claim 3, wherein the manifold alignment is performed using low rank alignment.

6. The apparatus of claim 5, wherein the measurements of the sensor are related to the relative mechanical movement of the robot during operational movement of the robot, and wherein the diagnostic device is in information communication with a transceiver configured to receive the measurements from the sensor.

7. The apparatus of claim 5, wherein the distance is calculated using an L2 norm measurement, and wherein degraded health of the robot can be determined by comparing the distance calculated by an L2 norm to a standard.

8. An apparatus, comprising:

a robotic diagnostic device comprising at least one input and one output and configured to assist in determining a health of a robot having a moving mechanical component by distinguishing operational changes of the robot from health changes of the robot, the robotic diagnostic device configured to: aligning data features of the training data and data features of the operational data in a common subspace; and utilizing a classifier trained on the training data and applied within the common subspace of the operational data, wherein the operational data comprises features related to mechanical movements of the robot, the mechanical movements targeting a health assessment, and the training data comprises features related to reference mechanical movements to which the operational data is to be compared.

9. The apparatus of claim 8, wherein the robotic diagnostic device comprises a computer processing device and a computer memory, wherein the memory comprises instructions that, when executed by the computer processing device, operate to provide an output signal indicative of the change in health of the robot, and wherein the robotic diagnostic device is configured to calculate a distance such that the distance is responsive to the change in health of the robot and is insensitive to changes in operation of the robot.

10. The apparatus of claim 8, wherein the classifier is a support vector machine.

11. The apparatus of claim 10, wherein the robotic diagnostic device is configured to: domain adaptation is used to align features of the training data and features of the operational data in a common subspace.

12. The apparatus of claim 8, wherein the robotic diagnostic device is configured to use unsupervised transfer learning.

13. The device of claim 12, wherein the distance between the training data and the operational data within the common subspace is calculated using an L2 norm.

14. The apparatus of claim 8, further comprising a robot coupled to the robotic diagnostic device, the robot having a movable component and a sensor configured to acquire data related to movement of the movable component, and wherein the robotic diagnostic device is configured to set an indication when the distance exceeds a health criterion.

15. A method, comprising:

effecting relative movement of the first robot movable member with respect to the second robot movable member;

collecting data relating to the relative movement;

extracting moving features of the data such that the moving features can be compared to training features;

projecting at least one of the moving features and the training features onto a common subspace including both the moving features and the training features to form estimated moving features and estimated training features; and

a classifier is utilized that is trained on the projected training features and applied to the projected moving features.

16. The method of claim 15, further comprising: comparing the distance to a health criterion, and setting an indication if the distance exceeds the health criterion.

17. The method of claim 16, wherein the determining comprises: an L2 norm is calculated between the estimated movement features and the estimated training features.

18. The method of claim 17, wherein utilizing a classifier comprises: a support vector machine is used.

19. The method of claim 17, wherein the projecting comprises: unsupervised alignment of the moving features and the training features.

20. The method of claim 19, further comprising: setting an indication when the distance exceeds a threshold.

Technical Field

The present invention relates generally to condition monitoring and in particular, but not exclusively, to operation independent robot condition monitoring.

Background

Current machine condition monitoring algorithms are operation dependent, requiring that current robot operations with a set of operational movements (such as those associated with the production environment) be the same as the movements of the robot operations of the robot mathematical model used to establish the health baseline, for comparison purposes. This requirement for similar movements may prevent efficient operation, as the robot's operational movements in the ongoing production environment must be stopped (at least when the current robot operation is not the same as the operation used to establish the baseline) to allow comparison with the healthy baseline model. If the operator chooses to forgo maintenance monitoring in order to increase the uptime (uptime), problems may arise, leading in the worst case to a breakdown of the robot. Therefore, there remains a need for further contributions in this area of technology.

Disclosure of Invention

One embodiment of the present invention is a unique condition monitoring device. Other embodiments include apparatuses, systems, devices, hardware, methods, and combinations for monitoring the condition of a robot independent of the particular robot operation. Further embodiments, forms, features, aspects, benefits, and advantages of the present application will become apparent from the description and drawings provided herein.

Drawings

Fig. 1 shows one embodiment of a robot.

FIG. 2 illustrates an embodiment of a computing device.

Fig. 3 illustrates a health monitoring assessment method.

Fig. 4 illustrates a health monitoring assessment method.

Fig. 5 shows a comparison of data using the techniques described herein.

Fig. 6 shows a comparison of data using the techniques described herein.

Detailed Description

For the purposes of promoting an understanding of the principles of the invention, reference will now be made to the embodiments illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the invention is thereby intended. Any alterations and further modifications in the described embodiments, and any further applications of the principles of the invention as described herein are contemplated as would normally occur to one skilled in the art to which the invention relates.

Referring to fig. 1, a schematic diagram of a robot 50 includes a plurality of movable robotic components 52 that may be used in the robot to perform an action. In the illustrated embodiment, robot 50 is depicted as interacting with target 54 using an effector 56, but it will be understood that the robot may assume other roles that may or may not involve physical interaction with objects, such as those roles for which robot 50 is limited to viewing only. As shown in fig. 1, the robot 50 may be mounted on a fixed base, but other forms are also contemplated, such as forms involving a mobile robot. The robotic component 52 may take any of a variety of forms that can be used to position the actuator 56, such as an arm, a link, a beam, and so forth. The robot 50 may include any number of movable components 52, and the movable components 52 may take on different sizes, shapes, and other characteristics. Further, the components 52 may be interconnected to one another by any of a variety of useful mechanisms, such as linkages and/or gears 58 (to illustrate two examples only). The member 52 may be actuated via any suitable actuation device 60, such as an electrical actuator, a pneumatic or hydraulic piston, an electromechanical actuator, or the like. The actuator 56 may take any of a variety of forms, such as a gripper, suction actuator, belt, and the like. As will be appreciated, the sensors 62 may be used to detect an operating condition of the robot 50. Such sensors 62 may be part of robot 50, may be coupled to robot 50, or may be remote from robot 50. For the purpose of illustrating only some non-limiting examples, sensors 62 may be associated with the actuators 60 for moving the robot, such as current sensors, hydraulic pressure sensors, and the like. In additional and/or alternative forms, the sensor 62 may be a stand-alone sensor for detecting the position, orientation, operating conditions, etc. of the robot. For example, the vibration sensor 62 may be used to sense a condition of the robot. In still additional and/or alternative forms, sensor 62 may be a camera for detecting electromagnetic conditions of robot 50.

Robot 50 may be coupled to a computing device 55, where computing device 55 may be local to robot 50 or located at a remote location. Such computing devices 55 may be used to control, observe, and/or estimate the condition of the robot 50, among other potential uses. As will be described further below, embodiments of computing device 55 may be used to detect the relative health of robot 50. It should be understood that embodiments may also be applied to other mechanically movable devices, whether of the robotic type or not.

Turning now to fig. 2, and with continued reference to fig. 1, a schematic diagram of a computing device 55 for controlling, observing, estimating a condition and/or assessing the health of the robot 50 is depicted. Computing device 55 includes a processing device 64, an input/output device 66, a memory 68, and operating logic 70. Further, the computing device 55 may be configured to communicate with one or more external devices 72.

Input/output device 66 may be any type of device that allows computing device 55 to communicate with external device 72. For example, the input/output device may be a network adapter, an RF device, a network card, or a port (e.g., a USB port, a serial port, a parallel port, VGA, DVI, HDMI, firewire, CAT 5, or any other type of port). The input/output devices 66 may include hardware, software, and/or firmware. It is contemplated that the input/output device 66 includes more than one of these adapters, cards, or ports.

External device 72 may be any type of device that allows data to be input or output from computing device 55. In one non-limiting example, the external device 72 is one or more of the sensors 62. For the purpose of illustrating only a few additional non-limiting examples, the external device 72 may be another computing device, a server, a printer, a display, an alarm, an illuminated indicator, a keyboard, a mouse button, or a touch screen display. Further, it is contemplated that the external device 72 may be integrated into the computing device 55. For example, the computing device 55 may be a smartphone, laptop computer, or tablet computer. It is also contemplated that there may be more than one external device in communication with computing device 55. The external device may be co-located with computing device 55 or alternatively located remotely from the computer.

The processing device 64 may be of a programmable type, a dedicated hardwired state machine, or a combination thereof; and may also include a plurality of processors, Arithmetic Logic Units (ALUs), Central Processing Units (CPUs), and the like. For forms of processing device 64 having multiple processing units, distributed, pipelined, and/or parallel processing may be suitably utilized. Processing device 64 may be dedicated to performing only the operations described herein or may be utilized in one or more additional applications. In the depicted form, the processing device 64 is of a programmable type that performs algorithms and processes data according to operating logic 70 defined by programming instructions (such as software or firmware) stored in memory 68. Alternatively or additionally, the operating logic 70 for the processing device 64 is defined at least in part by hardwired logic or other hardware. Processing device 64 may be comprised of one or more components of any type suitable for processing signals received from input/output device 66 or elsewhere, and providing desired output signals. Such components may include digital circuitry, analog circuitry, or a combination of both.

The memory 68 may be of one or more types, such as solid state, electromagnetic, optical, or a combination of these. Further, the memory 68 may be volatile, non-volatile, or a hybrid of these types, and some or all of the memory 68 may be of a portable type, such as a diskette, a tape, a memory stick, a tape cartridge, or the like. Additionally, memory 68 may store data that is manipulated by operating logic 70 of processing device 64, such as data representing signals received from and/or sent to input/output device 66, as an addition to, or in place of, storing programming instructions that define operating logic 70, to name one example. As shown in fig. 2, memory 68 may be included in processing device 64 and/or coupled to processing device 64.

The operating logic 70 may include the algorithms and steps of the controller, whether the controller includes the entire suite of algorithms necessary to affect the movements and actions of the robot 50, or whether the computing device 55 includes only the algorithms necessary to assess the health of the robot 50. The operating logic may be stored in a memory device, whether of the volatile or nonvolatile type, and may be expressed in any suitable type, such as but not limited to source code, object code, and machine code.

Turning now to FIG. 3, one embodiment of a system for detecting the health of a robot 50 is shown. Monitoring the health of the robot may be important in preventing accidental robot failures and crashes. Monitoring the condition of, and/or the interaction of, any of the robotic components 52, actuation devices 60, sensors 62 helps determine the overall health of the robot 50. In only one non-limiting example, monitoring for backlash can be used to assess the health of the robot. To observe degradation and predict failure, the model may be trained using signals obtained from robot 50 (or a similar robot used for this purpose) while the robot is in a healthy or baseline state. The training phase associated with modeling is generally comprised of data collection 74 (data collection 74 may include some amount of pre-processing), feature extraction 76, data comparison 78, and decision making step 80, decision step 80 differentiating data corresponding to healthy or abnormal conditions. One or more computing devices 55 may be used in any of data collection 74, feature extraction 76, data alignment 78 in a common environment, comparison of data, and decision making 80. As will also be appreciated, the sensors 62 may be used to measure relevant details and provide data for the data collection 74.

Once the model has been trained using a healthy/baseline robot dataset (sometimes also referred to as a training dataset), signals obtained from the robot in later operations may be compared to the trained model, but care must be taken. As described herein, the robot may undergo a first set of operations characterized by a particular sequence of movements of the robot component 52 (and/or the robot effector 56). The robot may also be subjected to a second set of operations characterized by different sequences of movements of the robot component 52 (and/or the robot effector 56) in the same or different combinations. Thus, the operation may be the same as the training data, but not always the same. If the features determined from the operation of the robot deviate significantly from the features present in the model, the robot may be considered to be in an abnormal state and future faults may be predicted, but care must be taken to ignore differences that are due solely to operational changes and not health changes. The embodiments described below may be used to help compare operational data to health/baseline data for a robot, regardless of whether the operational movements of the robot are the same as those used to evaluate the health/baseline robot. The embodiments described herein introduce a manifold alignment step after feature extraction 76 and before making the final decision 78.

One technique for extracting relevant features from data collected via sensors 62 involves the use of principal component analysis. Another technique for extracting features in step 76 prior to manifold alignment employs the use of a Short Time Fourier Transform (STFT) of the robot signal, and the combination is used for feature extraction. The following references to principal component analysis are merely examples and are not intended to be limiting, as the STFT method may also be used to extract features. Thus, the term "features" as referred to herein may refer to features extracted using STFT as well as principal component analysis. As will be appreciated, principal component analysis may be performed on the data set, whether the data is representative of raw or calculated data, and/or representative of data collected from the sensors, further processed/calculated and/or used in the estimation routine by itself or in conjunction with other data. Let v₁、v₂、…、v_PRepresenting features X ∈ R extracted from the training signal^NXPThe main component (c). These principal components (vectors) span a subspace, which represents the training data. Now let Y ∈ R^NXPFor extracting the main component U from the test data₁、U₂，…，U_PThe characteristics of (1). If X and Y are features from two different data sets with the same robot operation, v₁、v₂、…、v_PX and Y can be represented and Y can be compared to X in this subspace. Thus, a deviation of Y from X is an indicator of an anomaly.

If X and Y are correspondingAt two different robot operating characteristics, then v₁、v₂、…、v_PY may not be represented. Thus, Y may be detected as abnormal, even though it may be normal behavior. In the embodiments of the present application herein, the subspace of X and Y is considered to be two points on the manifold. It is assumed that the subspace representing the dataset moves in the manifold as the operation changes. To compensate for this movement, the features of the two data sets may be aligned prior to comparison. Manifold alignment is used in the alignment by embedding each input set into a shared potential manifold space when it is assumed that the data sets share a common infrastructure. The techniques described herein find a common subspace between test and training data sets through unsupervised techniques (such as manifold alignment), which in some embodiments may include solving optimization problems.

FIG. 3 depicts an embodiment in which an optimization problem is solved to help determine a common subspace of test and training data. Non-limiting techniques to determine the common subspace include Domain Adaptation (Domain Adaptation), which may include manifold alignment, Local Linear Embedding (LLE), Low Rank Embedding (LRE), and Low Rank Alignment (LRA), to name a few. Features of one of the data sets may also be transformed into a subspace of another of the data sets. LRA is a variant of LRE, and can be used to find v₁、v₂、…、v_PAnd u₁、u₂、…、u_PA common subspace in between (such as features extracted using STFT). One such method for determining LRA is described in the paper "Aligning Mixed Manifolds" by Boucher, Carey, Mahadevan and Dyar, the contents of which are incorporated herein by reference.

The method of determining the LRA is as follows. To begin alignment, Singular Value Decomposition (SVD) is used to decompose X, X ═ USV^TColumns of V and S

Root I₂＝{i：s_i≤1∨s_iE S, is set and is divided into V ═ V₁V₂]And S ═ S₁S₂]. For in LRE

The optimal closed form solution of the loss function used in the method is

Then R (x), R (y) are calculated independently. In one form, the computations may be performed in parallel to reduce computation time. The block matrix R, C ∈ R^NXNIs defined as:

and F ∈ R^N×dIs defined as:

the second step of the LRA is to calculate X, Y the embedded F by minimizing the loss function:

where μ ∈ [0,1] is a hyperparameter that controls the importance of correspondence between sets. The first term in the above equation describes the local geometry within each of the data sets, and the second term in the above equation describes the correspondence between the sets. The loss function can be reduced to the sum of the matrix traces as follows:

in a similar manner to LLE and LRE, constraint F may be introduced^TF ═ I to ensure loss function

Is minimized toAnd (4) solving the problem.

Thus:

wherein, M ═ R (I-R)^T(I-R) to construct the loss function from the above equation, take the right side and introduce the lagrangian multiplier Λ,

to minimize the above equation, the root of its partial derivative is found:

from the above system of equations, we leave the matrix eigenvalue problem as follows:

to solve the above minimization problem, the d-minimum non-zero eigenvector of the following matrix is calculated: (1-. mu.) M + 2. mu.L.

Because the matrix M + L is guaranteed to be symmetric, semi-Positive (PSD) and sparse, the intrinsic problem can be solved efficiently. These specific properties are derived from

Wherein the preparation method comprises the steps of through construction,

is a PSD diagonal matrix and C^(X,Y)Is a sparse matrix.

The algorithm for computing low rank alignment is as follows:

it should be appreciated that once the data set has been projected onto the common subspace, a comparison may be made between the health/baseline training data and the operational data. A classifier may be used merely to illustrate some non-limiting examples to facilitate the comparison. Classifiers such as, but not limited to, support vector machines and various distance techniques such as the L2 norm may be used.

Fig. 5 and 6 illustrate several results using the techniques described herein. Fig. 5 shows a comparison of data using manifold alignment versus not using manifold alignment when there is a change in operation of only robot 50. The top curve of fig. 5 shows the resulting difference if the operation of the robot is not the same as the operation used to perform the training, and also represents the resulting difference if the principal component analysis results are not projected onto the common subspace. The operational changes detected as anomalies when manifold alignment is not used are shown in the top curve of fig. 5, and where principal component analysis is used on the STFT output. Note the large L2 norm of the difference between X and Y. The bottom curve in fig. 5 shows the results using STFT and shows therein the L2 norm of the difference between the two data sets X and Y after projection onto the common subspace using manifold alignment. Note the relatively small scale of the difference between X and Y in the bottom curve of fig. 5 and the lack of sensitivity to operational changes.

Fig. 6 depicts another experiment, this time one that did not use operational changes, but introduced robot faults. The top curve of fig. 6 is again used to illustrate the L2 norm, where the data sets are not projected onto a common subspace, and where principal component analysis is used on the STFT output. The bottom curve of fig. 6 illustrates the results using STFT, and where the L2 norm of the dataset is projected onto the common subspace using manifold alignment. Fig. 6 illustrates that both techniques can be used to determine robot faults. Combining the results of fig. 5 and 6, it can be seen that the techniques herein for projecting a data set onto a common subspace are relatively insensitive to robot operation variations and are still sensitive to detecting robot faults.

Returning now to fig. 2 and 3, and with continued reference to fig. 1, various other aspects of how data is collected, calculated, shared, and acted upon will also be described. Both the robot 50 and the computing device 55 (referred to as diagnostic devices) implementing the subspace projection/alignment techniques described herein may be located locally relative to one another, although such a configuration is not a requirement. In one form, data relating to the operation of the robot (e.g., via sensors 62) may be transmitted directly to the diagnostic device, whether located locally or remotely from one another. For example, data may be transmitted via wired and/or wireless techniques that may employ any type of data format (e.g., packet transmission, pulse width modulation, etc.). In another alternative and/or additional form, the data from the sensors 62 may be delivered to a data cloud to be accessed by the diagnostic device. The transmitted data may represent raw data collected by the sensors, but in some embodiments the data may represent measurements that have been pre-processed (e.g., conversion from meter counts to units of measure, etc.) and/or further calculated, processed, etc. prior to delivery to the diagnostic device, the data may be processed by a single computing device 55 to determine the health of the robot 50, but in other forms the computing device 55 may be a distributed computing device, with different calculations/analyses/steps being performed by various devices dispersed over distributed resources. It should be understood that the diagnostic device may reside on the robot in some forms, and may be remote in other forms.

The data computed from the classifier (e.g., regression calculations (e.g., the L2 norm)) may be shared back to the local region of the robot after the dataset has been projected into the common subspace, it may be used in a database of examinations, and/or it may be stored for later evaluation, as well as other potential uses. The techniques herein may utilize criteria that may be used to identify robot faults. For purposes of merely illustrating one non-limiting example, the absolute value of the distance calculated by the L2 norm may be used to define a standard that defines an unhealthy robot. In other forms, the relative value calculated from the L2 norm may also be used as a criterion, merely to illustrate some non-limiting examples, such as a deviation from a value such as a mean. Other similar or different determinations may be made for other classifiers.

An indication may be set if the data computed from the classifier exceeds a criterion (e.g., an indication may be set if the absolute/relative/etc. of the distance (e.g., L2 norm) exceeds a criterion). Such indications may include text warnings entered into a log, visual warnings on a computer screen, audible warnings in a local area of the robot, warning lights, etc., such indications may be general warnings that generally inspect the robot, while other indications may be specific to replacing or repairing a particular robot component. In short, any number of passive and active techniques may be used to automatically mark the robot's unhealthy state.

One aspect of the present application includes an apparatus comprising: a robot having components configured to undergo relative mechanical movement during operation of the robot; a sensor coupled to the robot configured to detect an operating condition of a component; a diagnostic device configured to receive measurements from the sensors and including a memory having training data, the diagnostic device configured to distinguish between changes in the health of the robot and changes in the operation of the robot, and having computer-based instructions, the diagnostic device configured to: the method includes computing a principal component analysis of the measurements from the sensors to provide operational data, determining a common subspace between the training data and the operational data, and utilizing a classifier trained on the training data and applied to the operational data in the common subspace.

Features of the present application provide wherein the diagnostic device comprises a computer processing device and a computer memory, wherein the memory comprises computer-based instructions that when executed by the computer processing device operate to provide an output signal indicative of a distance between training data and operational data in a common subspace, the output signal having greater sensitivity to changes in the health of the robot and less sensitivity to changes in the operation of the robot.

Another feature of the present application provides wherein the diagnostic device is further configured to determine the common subspace via unsupervised domain adaptation.

Yet another feature of the present application provides wherein the diagnostic device is further configured to determine the common subspace via unsupervised domain adaptation.

Yet another feature of the present application provides wherein the unsupervised domain adaptation is performed using manifold alignment.

Yet another feature of the present application provides wherein the manifold alignment is performed using low rank alignment.

Yet another feature of the present application provides wherein the measurements of the sensor during the operational movement of the robot are related to the relative mechanical movement of the robot, and wherein the diagnostic device is in information communication with a transceiver configured to receive the measurements from the sensor

Yet another feature of the present application provides wherein the distance is calculated using an L2 norm measurement, and wherein the degraded health of the robot may be determined by comparing the L2 norm calculated distance to a standard.

Another aspect of the present application provides an apparatus comprising: a robotic diagnostic device including at least one input and one output and configured to assist in determining a health of a robot having moving mechanical parts by distinguishing operational changes of the robot from health changes of the robot, the robotic diagnostic device configured to align data features of training data and data features of the operational data in a common subspace, and utilize a classifier trained on the training data and applied within the common subspace of the operational data, wherein the operational data includes features related to mechanical movements of the robot, the mechanical movements are targeted for health assessment, and the training data includes features related to reference mechanical movements to which the operational data is to be compared.

Features of the present application provide wherein the robotic diagnostic device includes a computer processing device and a computer memory, wherein the memory includes instructions that, when executed by the computer processing device, operate to provide an output signal indicative of a change in health of the robot, and wherein the robotic diagnostic device is configured to calculate the distance such that the distance is responsive to the change in health of the robot and insensitive to changes in operation of the robot.

Another feature of the present application provides wherein the classifier is a support vector machine.

Yet another feature of the present application provides wherein the robotic diagnostic device is configured to align features of the training data and features of the operational data in a common subspace using domain adaptation.

Yet another feature of the present application provides wherein the robotic diagnostic apparatus is configured to use unsupervised transfer learning.

Yet another feature of the present application provides wherein the distance between the training data and the operational data within the common subspace is calculated using an L2 norm.

Still another feature of the present application further includes a robot coupled to the robotic diagnostic device, the robot having a movable component and a sensor configured to acquire data related to movement of the movable component, and wherein the robotic diagnostic device is configured to set an indication when the distance exceeds a health criterion.

Yet another aspect of the present application provides a method comprising: effecting relative movement of the first robot movable member with respect to the second robot movable member; collecting data relating to the relative movement; extracting the moving features of the data so that the moving features can be compared with the training features; projecting at least one of the moving features and the training features onto a common subspace including both the moving features and the training features to form estimated moving features and estimated training features; and utilizing a classifier trained on the projected training features and applied to the projected moving features.

The method further comprises comparing the distance to a health criterion and setting an indication in case the distance exceeds the health criterion.

Another feature of the present application provides wherein the determining comprises calculating an L2 norm between the evaluation movement feature and the evaluation training feature.

Still another feature of the present application provides for utilizing the classifier includes using a support vector machine.

Yet another feature of the present application provides for the unsupervised alignment wherein the projection includes moving features and training features.

Yet another feature of the present application further includes setting an indication when the distance exceeds a threshold.

While the invention has been illustrated and described in detail in the drawings and foregoing description, the same is to be considered as illustrative and not restrictive in character, it being understood that only the preferred embodiment has been shown and described and that all changes and modifications that come within the spirit of the invention are desired to be protected. It should be understood that while the use of words such as preferred, preferably, preferred or more preferred in the description above indicate that the feature so described may be more desirable, it nonetheless may not be necessary and embodiments lacking the same may be contemplated as within the scope of the invention, the scope being defined by the claims that follow. In reading the claims, it is intended that when words such as "a," "an," "at least one," or "at least a portion" are used, the claims be limited to only one item unless specifically stated to the contrary in the claims. When the term "at least a portion" and/or "a portion" is used, the item can include a portion and/or the entire item unless clearly indicated to the contrary. Unless specified or limited otherwise, the terms "mounted," "connected," "supported," and "coupled" and variations thereof are used broadly and encompass both direct and indirect mountings, connections, supports, and couplings. Further, "connected" and "coupled" are not restricted to physical or mechanical connections or couplings.