阅读说明：本技术 一种基于人工智能的挤塑设备故障预警方法及系统 (Artificial intelligence-based extrusion molding equipment fault early warning method and system ) 是由 熊莲 王海俊 于 2021-10-28 设计创作，主要内容包括：本发明涉及人工智能技术领域,具体涉及一种基于人工智能的挤塑设备故障预警方法及系统,包括获取用于检测挤塑设备故障所需的设定时间段以及预先训练好的挤塑设备故障类型预测网络；实时获取所述设定时间段内待检测挤塑设备在每个时刻的运行日志数据,并根据所述设定时间段内待检测挤塑设备在每个时刻的运行日志数据和预先训练好的挤塑设备故障类型预测网络,预测待检测挤塑设备是否发生故障以及在发生故障时对应的故障类别；若预测到待检测挤塑设备发生故障,则根据故障类别进行预警。本发明可在故障发生之前,对设备可能即将出现的故障进行预警,避免了设备出现故障,提高了制品的合格率。(The invention relates to the technical field of artificial intelligence, in particular to a fault early warning method and a fault early warning system for extrusion molding equipment based on artificial intelligence, which comprises the steps of obtaining a set time period required for detecting faults of the extrusion molding equipment and a pre-trained fault type prediction network of the extrusion molding equipment; acquiring the running log data of the extrusion equipment to be detected at each moment in the set time period in real time, and predicting whether the extrusion equipment to be detected fails or not and the corresponding fault category when the extrusion equipment to be detected fails according to the running log data of the extrusion equipment to be detected at each moment in the set time period and a pre-trained extrusion equipment fault type prediction network; and if the extrusion molding equipment to be detected has faults, early warning is carried out according to the fault type. The invention can early warn the possible impending failure of the equipment before the failure occurs, thereby avoiding the failure of the equipment and improving the qualification rate of products.)

1. An artificial intelligence-based extrusion molding equipment fault early warning method is characterized by comprising the following steps:

acquiring a set time period required for detecting faults of extrusion molding equipment and a pre-trained extrusion molding equipment fault type prediction network;

acquiring the running log data of the extrusion equipment to be detected at each moment in the set time period in real time, and predicting whether the extrusion equipment to be detected fails or not and the corresponding fault category when the extrusion equipment to be detected fails according to the running log data of the extrusion equipment to be detected at each moment in the set time period and a pre-trained extrusion equipment fault type prediction network;

and if the extrusion molding equipment to be detected has faults, early warning is carried out according to the fault type.

2. The artificial intelligence based extrusion apparatus fault early warning method as claimed in claim 1, wherein the determination of the set period of time required for detecting the fault of the extrusion apparatus includes:

acquiring running log text data and monitoring image data at T moments in the historical working process of extrusion molding equipment;

obtaining embedded vectors corresponding to the running log text data at the T moments according to the running log text data at the T moments;

constructing an undirected graph structure according to the embedded vectors corresponding to the running log text data at the T moments;

and determining a set time period required for detecting the faults of the extrusion molding equipment according to the undirected graph structure and the monitoring image data at T moments.

3. The artificial intelligence based extrusion molding equipment fault early warning method according to claim 2, wherein the step of constructing an undirected graph structure according to the embedded vectors corresponding to the run log text data at the T moments comprises:

respectively taking the embedded vectors corresponding to the running log text data at T moments as a node, and calculating the edge weight between any two nodes according to the embedded vector and the moment corresponding to each node;

and obtaining an undirected graph structure according to all the nodes and the edge weights between any two nodes.

4. The artificial intelligence based extrusion molding equipment fault early warning method according to claim 3, wherein the calculation formula of the edge weight between any two nodes is as follows:

wherein the content of the first and second substances,the edge weight value between the node corresponding to the embedded vector corresponding to the execution log text data at the time a and the node corresponding to the embedded vector corresponding to the execution log text data at the time b,for the embedded vector corresponding to the run log text data at time a,and the embedded vector corresponding to the running log text data at the moment b.

5. The artificial intelligence based extrusion molding apparatus malfunction early warning method according to claim 2, wherein the step of determining the set time period required for detecting the malfunction of the extrusion molding apparatus based on the undirected graph structure and the monitoring image data at T times comprises:

determining a set time period required for detecting the faults of the extrusion molding equipment according to the undirected graph structure and the monitoring image data at T moments, and constructing a loss function of a graph cut algorithm;

determining an initial graph cut number, and segmenting an undirected graph structure by utilizing a graph cut algorithm according to the initial graph cut number and a loss function so as to determine a graph segmentation mode and a loss function value corresponding to the initial graph cut number;

updating the initial graph cut number, and segmenting the undirected graph structure by utilizing a graph cut algorithm according to the updated graph cut number and the loss function so as to determine a graph segmentation mode and a loss function value corresponding to the updated graph cut number;

determining whether the graph cut number updating stopping condition is met or not according to the loss function value corresponding to the initial graph cut number and the loss function value corresponding to the updated graph cut number, and if the graph cut number updating stopping condition is not met, continuously updating the current updated graph cut number until the graph cut number updating stopping condition is met;

and determining the set time period required for detecting the faults of the extrusion equipment according to the graph segmentation mode corresponding to the graph segmentation number after the last updating.

6. The artificial intelligence based extrusion molding equipment fault pre-warning method according to claim 5, wherein the step of determining the set time period required for detecting the extrusion molding equipment fault according to the graph segmentation mode corresponding to the number of graph segments updated at the last time comprises:

determining the number of nodes of each divided sub-graph according to the graph division mode corresponding to the graph division number after the last updating;

and calculating the difference value of the time corresponding to any two nodes in all the subgraphs except the subgraph corresponding to the maximum node number, and taking the minimum value of all the difference values as the set time period required for detecting the faults of the extrusion molding equipment.

7. An artificial intelligence based extrusion apparatus fault pre-warning method as claimed in claim 5, wherein the corresponding calculation formula of the loss function of the graph cut algorithm is:

wherein the content of the first and second substances,the corresponding energy loss function when the graph is divided into L,the corresponding region penalty item when the graph division mode is L,the corresponding boundary penalty term when the graph is divided into L,and the corresponding monitoring image information entropy is obtained when the graph segmentation mode is L.

8. The artificial intelligence based extrusion molding equipment fault early warning method according to claim 7, wherein the corresponding calculation formula of the monitoring image information entropy when the graph segmentation mode is L is as follows:

wherein the content of the first and second substances,the corresponding monitoring image information entropy when the graph segmentation mode is L,for the monitoring image at time a in the k-th sub-image that is segmented,for the monitoring image at time b in the k-th sub-image that is segmented,for the number of nodes in the k-th sub-graph that is partitioned,the corresponding number of subgraphs when the graph division mode is L.

9. An artificial intelligence based extrusion device fault early warning system comprising a memory and a processor and a computer program stored in the memory and running on the processor, the processor being coupled to the memory, the processor, when executing the computer program, implementing the artificial intelligence based extrusion device fault early warning method as claimed in any one of claims 1 to 8.

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a fault early warning method and system for extrusion molding equipment based on artificial intelligence.

Background

The extrusion molding apparatus refers to a manufacturing apparatus applied to an extrusion molding process. Extrusion is one of the most important forms of plastic material processing, which is suitable for most plastic materials except some thermosetting plastics, and about 50% of thermoplastic articles are completed by extrusion, and extrusion is also used in a large number for the formation of chemical fibers and thermoplastic elastomers and rubber articles. Extrusion molding is a molding method in which a polymer material melted by heating is forced to pass through a die of a machine head under the pushing of pressure by means of the extrusion action of a screw or a plunger to form a continuous profile with a constant cross section. The extrusion forming process mainly comprises the processes of feeding, melting and plasticizing, extrusion forming, shaping, cooling and the like.

The material temperature of the extrusion molding equipment fluctuates, thereby affecting the quality of products, the strength of each point of the products is different, residual stress is generated, and the surface is dark and dull. Factors that influence extrusion rate include: head resistance, screw and barrel configuration, screw speed, heating, cooling systems, plastic properties, and therefore, extrusion rate also fluctuates, affecting the geometry and size of the article. Therefore, temperature, pressure, and extrusion rate all fluctuate. The existing fault monitoring mode is mainly characterized in that after a fault occurs, the extrusion molding equipment is manually controlled to stop running. The monitoring mode has low efficiency, wastes human resources, cannot realize early warning of faults, cannot avoid defective products and has low product qualification rate.

Disclosure of Invention

The invention aims to provide an artificial intelligence-based extrusion molding equipment fault early warning method and system, which are used for solving the problem that the qualification rate of products is low because the existing extrusion molding equipment fault monitoring mode cannot early warn in advance.

In order to solve the technical problem, the invention provides an artificial intelligence-based extrusion molding equipment fault early warning method, which comprises the following steps of:

acquiring a set time period required for detecting faults of extrusion molding equipment and a pre-trained extrusion molding equipment fault type prediction network;

acquiring the running log data of the extrusion equipment to be detected at each moment in the set time period in real time, and predicting whether the extrusion equipment to be detected fails or not and the corresponding fault category when the extrusion equipment to be detected fails according to the running log data of the extrusion equipment to be detected at each moment in the set time period and a pre-trained extrusion equipment fault type prediction network;

and if the extrusion molding equipment to be detected has faults, early warning is carried out according to the fault type.

Further, the determination of the set period of time required for detecting a failure of the extrusion apparatus includes:

acquiring running log text data and monitoring image data at T moments in the historical working process of extrusion molding equipment;

obtaining embedded vectors corresponding to the running log text data at the T moments according to the running log text data at the T moments;

constructing an undirected graph structure according to the embedded vectors corresponding to the running log text data at the T moments;

and determining a set time period required for detecting the faults of the extrusion molding equipment according to the undirected graph structure and the monitoring image data at T moments.

Further, the step of constructing an undirected graph structure according to the embedded vectors corresponding to the running log text data at the T moments includes:

respectively taking the embedded vectors corresponding to the running log text data at T moments as a node, and calculating the edge weight between any two nodes according to the embedded vector and the moment corresponding to each node;

and obtaining an undirected graph structure according to all the nodes and the edge weights between any two nodes.

Further, the calculation formula of the edge weight between any two nodes is as follows:

wherein the content of the first and second substances,the edge weight value between the node corresponding to the embedded vector corresponding to the execution log text data at the time a and the node corresponding to the embedded vector corresponding to the execution log text data at the time b,embedding corresponding to running log text data for time aThe amount of the compound (A) is,and the embedded vector corresponding to the running log text data at the moment b.

Further, the step of determining a set time period required for detecting the malfunction of the extrusion apparatus based on the undirected graph structure and the monitoring image data at T times includes:

determining a set time period required for detecting the faults of the extrusion molding equipment according to the undirected graph structure and the monitoring image data at T moments, and constructing a loss function of a graph cut algorithm;

determining an initial graph cut number, and segmenting an undirected graph structure by utilizing a graph cut algorithm according to the initial graph cut number and a loss function so as to determine a graph segmentation mode and a loss function value corresponding to the initial graph cut number;

updating the initial graph cut number, and segmenting the undirected graph structure by utilizing a graph cut algorithm according to the updated graph cut number and the loss function so as to determine a graph segmentation mode and a loss function value corresponding to the updated graph cut number;

determining whether the graph cut number updating stopping condition is met or not according to the loss function value corresponding to the initial graph cut number and the loss function value corresponding to the updated graph cut number, and if the graph cut number updating stopping condition is not met, continuously updating the current updated graph cut number until the graph cut number updating stopping condition is met;

and determining the set time period required for detecting the faults of the extrusion equipment according to the graph segmentation mode corresponding to the graph segmentation number after the last updating.

Further, the step of determining the set time period required for detecting the failure of the extrusion molding equipment according to the graph division mode corresponding to the graph division number updated at the last time comprises the following steps:

determining the number of nodes of each divided sub-graph according to the graph division mode corresponding to the graph division number after the last updating;

and calculating the difference value of the time corresponding to any two nodes in all the subgraphs except the subgraph corresponding to the maximum node number, and taking the minimum value of all the difference values as the set time period required for detecting the faults of the extrusion molding equipment.

Further, the corresponding calculation formula of the loss function of the graph cut algorithm is as follows:

wherein the content of the first and second substances,the corresponding energy loss function when the graph is divided into L,the corresponding region penalty item when the graph division mode is L,the corresponding boundary penalty term when the graph is divided into L,and the corresponding monitoring image information entropy is obtained when the graph segmentation mode is L.

Further, when the graph division mode is L, the corresponding calculation formula of the monitoring image information entropy is as follows:

wherein the content of the first and second substances,the corresponding monitoring image information entropy when the graph segmentation mode is L,for the monitoring image at time a in the k-th sub-image that is segmented,for the monitoring image at time b in the k-th sub-image that is segmented,for the number of nodes in the k-th sub-graph that is partitioned,the corresponding number of subgraphs when the graph division mode is L.

In order to solve the technical problem, the invention further provides an artificial intelligence based extrusion molding equipment fault early warning system, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor is coupled with the memory, and the processor realizes the artificial intelligence based extrusion molding equipment fault early warning method when executing the computer program.

The invention has the following beneficial effects: according to the method, the set time period required for detecting the faults of the extrusion molding equipment is obtained, the operation log data of the corresponding extrusion molding equipment at each moment in the set time period is obtained in real time when the fault detection is carried out on the extrusion molding equipment, and the operation log data are input into the pre-trained extrusion molding equipment fault type prediction network, so that whether the extrusion molding equipment to be detected is in fault or not and the corresponding fault type when the fault occurs can be determined, and therefore early warning is carried out. According to the invention, the fault early warning of the extrusion molding equipment to be detected can be realized through the operation log data generated by the extrusion molding equipment to be detected, so that the equipment fault is avoided, and the qualification rate of products is effectively improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 is a flow chart of the fault early warning method of the extrusion molding equipment based on artificial intelligence.

Detailed Description

To further explain the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the embodiments, structures, features and effects of the technical solutions according to the present invention will be given with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

The embodiment provides an artificial intelligence-based extrusion molding equipment fault early warning method, and a corresponding flow chart is shown in fig. 1, and the method specifically comprises the following steps:

step S1: and acquiring a set time period required for detecting the faults of the extrusion molding equipment and a pre-trained prediction network of the fault type of the extrusion molding equipment.

Wherein the set time period required for detecting the failure of the extrusion apparatus is determined in advance, and the corresponding determination process comprises:

(1-1) acquiring running log text data and monitoring image data at T moments in the historical working process of the extrusion molding equipment.

In the historical working process of the extrusion molding equipment, running log text data and monitoring image data at different moments are obtained. The operation log text data refers to the text record in the historical operation process of the extrusion molding equipment, and the monitoring image data refers to the monitoring image of the material or product outlet after being enhanced by HDRTet.

In order to improve the accuracy of the results of the subsequent data analysis steps, the operating log text data and the monitoring image data in the historical working process of the extrusion molding equipment are acquired as much as possible, namely the total number T of the corresponding moments of the acquired data is large enough.

And (1-2) obtaining the embedded vector corresponding to the running log text data at the T moments according to the running log text data at the T moments.

Wherein, according to the data obtained in the step (1-1), two types of data can be obtained corresponding to each time point, namely each moment, wherein one type of data is running log text data, namely equipment running logs (text data), and the other type of data is monitoring image data, namely product monitoring videos.

Recording the device running log set acquired in the step (1-1) as NOTE, andwherein, in the step (A),represents the index of time points, T represents that the text data of the running log contains T time points in total,represents the firstAnd the equipment at each time point runs a log. The device operation log data includes: the equipment preset parameters (material heating target temperature t1, material cooling temperature t2, screw preset rotation speed w 1), and the equipment sensor readings (screw temperature t3, screw measurement rotation speed w2, material pressure p). Therefore, for the device running log at the ith time point,。

recording the monitoring image obtained in the step (1-1) as RECORD, andwherein i represents a time point index, T represents T time points corresponding to the equipment operation log in the monitoring image,representing the monitored image at the ith time point.

Since the data format of the device operation log set NOTE of the extrusion device is text data, the text data at each moment needs to be quantized for subsequent analysis of the quantized data. In the embodiment, word embedding is carried out by using Skip-gram, and the text record of each time point is mapped into a vector model. The purpose of this step is to quantify the text to facilitate subsequent processes to calculate and analyze the text data. Since Skip-grams are well known techniques, they are not described in detail here.

For the running log text data at each moment in the device running log set NOTE, an embedded vector of the text data at each moment can be obtained by using Skip-gram and recorded asWherein i represents the index of time points, T represents a total of T time points,representing an embedded vector with index i.

(1-3) constructing an undirected graph structure according to the embedded vectors corresponding to the running log text data at T moments, wherein the specific steps comprise:

(1-3-1) respectively taking the embedded vectors corresponding to the running log text data at T moments as a node, and calculating the edge weight between any two nodes according to the embedded vector and the moment corresponding to each node.

Through the steps (1-2), the text data at each time point has a unique embedded vector representation which is recorded asEmbedding vector of each time pointAs a node of the undirected graph structure, the undirected graph structure can be obtainedT nodes of (2). In this embodiment, the edge weights between nodes are related to the similarity of the embedded vectors, and the closer the temporal node distance is, the higher the correlation is. In order to avoid negative values of the Edge weights, the similarity between the embedded vectors is represented by cosine distances in this embodiment, and therefore, the Edge weight value between any two nodes in this embodiment is denoted as Edge:

wherein the content of the first and second substances,representing the edge weight between the node with index a and the node with index b, i.e. the edge weight between the node corresponding to the embedded vector corresponding to the running log text data at the time a and the node corresponding to the embedded vector corresponding to the running log text data at the time b,，the embedded vector with index a, namely the embedded vector corresponding to the running log text data of the moment a,the embedded vector with index b, namely the embedded vector corresponding to the running log text data of the moment b,is composed ofAndthe cosine of the distance of (a) is,representsThe resulting weights are calculated based on the time span. For theIf the time span is larger, the correlation is lower; and vice versa. In the present embodiment, it is preferred that,at this time, the following are obtained comprehensively:

(1-3-2) obtaining an undirected graph structure according to all nodes and the edge weights between any two nodes.

Based on the historical running log of the equipment, all nodes and the edge weights between any two nodes can be obtained, and then an undirected Graph structure Graph is constructed, wherein the embedded vector of each time point is the node of the Graph, and the edge weight between any two nodes with the index of a and the index of b is. So far, the undirected graph structure of the device history log is constructed.

(1-4) determining a set time period required for detecting a failure of the extrusion molding apparatus based on the undirected graph structure and the monitoring image data at T times.

In order to identify the failure point in the time series, the embodiment performs Graph segmentation by using a Graph cut algorithm, and further determines a set time period required for detecting the failure of the extrusion molding equipment, and the specific steps are as follows:

(1-4-1) determining a set time period required for detecting the failure of the extrusion molding equipment according to the undirected graph structure and the monitoring image data at T moments, and constructing a loss function of the graph cut algorithm.

The image segmentation algorithm is usually applied to the foreground and background segmentation of an image, and the original image segmentation algorithm is applied to the image segmentation, and target pixel points and background pixel points need to be defined, so that the intersection points of the original algorithm are divided into two types: source s and sink t. In the embodiment, the purpose of graph cutting is to distinguish the time node when the equipment is operating normally from the time node when the equipment may be in failure, so that the intersection points do not need to be classified. And the number S of the intersection points of the graph cut algorithm in the embodiment may be greater than 2 because the existing faults of the equipment may be diversified.

The original loss function of the graph cut algorithm is:

wherein the content of the first and second substances,the Graph is divided into L, namely the energy of Graph is divided into L;and represents a regional penalty item, when the nodes are correctly classified,minimum;the smaller the cut margin is for the corresponding region penalty term when the graph division mode is L,the smaller. However, in this embodiment, since the monitoring image data is also an important basis for determining whether a fault occurs, a penalty term of the monitoring image data needs to be added to the loss function of the graph cut algorithm, and the obtained loss function of the graph cut algorithm at this time is:

wherein the content of the first and second substances,the method is the corresponding monitoring image information entropy when the Graph segmentation mode is L, namely the information entropy of all video frames in the same sub-Graph after Graph is segmented. If the information entropy is smaller, the text data is accurately classified, and the monitoring image data is also accurately classified.

As can be known from the definition of graph cut algorithm, in the graph cut algorithm, the number S of the preset junction points represents the number of the subgraphs to be obtained by splitting. The manner and purpose of each penalty term is explained in detail below:

regional penalty termTypically, in the divided Graph, the sum of the similarity between each node and the preset junction should be the maximum, and then r (l) takes the minimum value at this time, the energy of the Graph is the minimum, in this embodiment,the calculation method of (c) is as follows:

wherein the content of the first and second substances,when the representative graph is divided into L, the region punishment item of the energy loss function; i represents the index of the node in Graph; k represents an index of a preset junction; k represents the number of subgraphs, K = S;representing the ith node vector;represents the kth intersection point vector (the vector is selected randomly);for the harmonic constant, the value of the harmonic constant should be 0.5, as a rule of thumb.

The boundary penalty term b (L) represents that, when Graph segmentation is performed in the segmentation method L, the smaller the weight of the broken edge is, the smaller the penalty is, and in this embodiment, the calculation method of b (L) is as follows:

wherein, a and b represent indexes of nodes;representing the edge weight between the node with index a and the node with index b.

Image penalty termWhen Graph segmentation is performed in the segmentation mode L, the euclidean distance between all the monitored image data in each sub-Graph is represented. Image penalty termThe smaller the number of images in the subgraph, the higher the similarity of all the images in the subgraph and the higher the clustering accuracy.

Wherein the content of the first and second substances,the corresponding monitoring image information entropy when the graph segmentation mode is L,the monitoring image with index number a in the k-th sub-image is divided, namely the monitoring image at the moment a in the k-th sub-image is divided,the monitoring image with index number b in the k-th sub-image is divided, namely the monitoring image at the moment b in the k-th sub-image is divided,for the number of nodes in the k-th sub-graph that is partitioned,the number of corresponding sub-graphs, i.e. the index of the predetermined intersection point, when the graph division mode is L.

(1-4-2) determining an initial graph cut number, and segmenting the undirected graph structure by utilizing a graph cut algorithm according to the initial graph cut number and the loss function, thereby determining a graph segmentation mode and a loss function value corresponding to the initial graph cut number.

Because the number of the fault categories is unknown, that is, the number K of the divided subgraphs is unknown, an optimized value K needs to be calculated, and a corresponding optimized division mode L is determined. In this embodiment, the number K of subgraphs is determined through a traversal method, first, the initial graph cut number K =2 is determined, K data nodes are randomly selected from the T data nodes as intersection points, and then an optimized graph cut method L in the current intersection point selection mode is obtained through a minimization loss function e (L). And then, re-selecting K junction points, calculating a loss function E (L), and circulating the steps until all possible junction point combination modes are completed by traversal. The intersection point selection mode with the minimum value of e (L) and the corresponding graph cut mode L are the optimized graph cut mode and the intersection point selection mode when the number of subgraphs K =2, the corresponding subgraph is the subgraph obtained by the optimized segmentation method, and the corresponding loss function value is recorded as the loss function value corresponding to the initial graph cut number K = 2.

(1-4-3) updating the initial graph cut number, and segmenting the undirected graph structure by using a graph cut algorithm according to the updated graph cut number and the loss function, thereby determining a graph segmentation mode and a loss function value corresponding to the updated graph cut number.

And updating the initial graph cut number, namely enabling the graph cut number K =3, and then determining a graph segmentation mode and a loss function value corresponding to the graph cut number K =3 according to the mode of the step (1-4-2).

(1-4-4) determining whether the graph cut number updating stopping condition is met according to the loss function value corresponding to the initial graph cut number and the loss function value corresponding to the updated graph cut number, and if the graph cut number updating stopping condition is not met, continuously updating the current updated graph cut number until the graph cut number updating stopping condition is met.

And comparing the loss function value corresponding to the initial graph cut number with the loss function value corresponding to the updated graph cut number, and judging that the graph cut number updating stopping condition is met when the difference value between the loss function value corresponding to the updated graph cut number and the loss function value corresponding to the initial graph cut number is smaller than a set difference threshold value, otherwise, judging that the graph cut number updating stopping condition is not met. In the present embodiment, the set difference threshold is set to 0.1. And if the condition that the graph cut number stops updating is not met, continuing updating the current updated graph cut number, namely enabling the graph cut number K =4, and repeating the steps (1-4-2) - (1-4-4) until the condition that the graph cut number stops updating is met. At this time, the graph cut number after the last update is the optimized intersection point number, the intersection point selection mode at this time is the optimized selection mode, and the graph cut mode L at this time is the optimized graph cut mode.

(1-4-5) determining a set time period required for detecting the failure of the extrusion molding equipment according to the graph division mode corresponding to the graph division number after the last update.

Through the steps (1-4-4), an optimized graph cutting mode can be obtained, the similarity between nodes in the subgraph after graph cutting is high enough, the similarity of video frames corresponding to the nodes in the subgraph is also high enough, and the similarity between the cut nodes is small enough. Assuming that M subgraphs are obtained by means of optimized graph cut, the nodes in each subgraph correspond to a class of data. Through logical reasoning, in the running process of the equipment, the time node of normal running is far longer than the time node of failure. Therefore, the subgraph with the largest number of nodes can represent the data set when the device operates normally. Other classes of nodes may characterize the point in time at which the device fails. Therefore, the node number of each divided sub-graph is determined according to the graph division mode corresponding to the graph division number after the last update, the difference value of the time corresponding to any two nodes in all other sub-graphs except the sub-graph corresponding to which the node number is the maximum is calculated, and the minimum value of all the difference values is used as a set time period required for detecting the extrusion molding equipment fault, namely, the time span between fault points, namely, the minimum fault interval time, which is denoted as MTBF.

Step S2: and acquiring the operation log data of the extrusion equipment to be detected at each moment in the set time period in real time, and predicting whether the extrusion equipment to be detected fails or not and the corresponding fault type when the extrusion equipment to be detected fails according to the operation log data of the extrusion equipment to be detected at each moment in the set time period and the pre-trained extrusion equipment fault type prediction network.

Through the above step S1, the time point at which the apparatus normally operates and the time point at which there is a failure have been accurately divided. For each type of fault, sequence data in an MTBF time period before each fault occurs and sequence data in an MTBF time period after each fault occurs can be selected, wherein the sequence data refers to operation log data of each time in the MTBF time period before each fault occurs and operation log data of each time in the MTBF time period after each fault occurs of equipment, and the operation log data are used as training set data of an extrusion molding equipment fault type prediction network constructed by an RNN network, the label of the operation log data corresponding to the equipment before each fault occurs is a fault type, and the label of the operation log data corresponding to the equipment after each fault occurs is that the equipment has no fault.

And training the extrusion molding equipment fault type prediction network by using the data set so as to obtain the trained extrusion molding equipment fault type prediction network. In the present embodiment, the activation function of the extrusion device failure type prediction network is a tanh function, and the loss function is a mean square error function. Since the specific training process for constructing the extrusion equipment fault type prediction network by using the RNN network and the extrusion equipment fault type prediction network belongs to the prior art, the detailed description is omitted here.

When fault detection is required to be carried out on the extrusion molding equipment, the operation log data of the extrusion molding equipment to be detected at each moment in a set time period MTBF is obtained in real time, word embedding operation is carried out on the operation log data of the extrusion molding equipment to be detected at each moment in the set time period MTBF, Skip-gram is still selected in the word embedding method, the embedding vector of the current time sequence is obtained and recorded as the embedding vector of the current time sequence. Will be provided withInputting the data into a failure type prediction network of extrusion molding equipment so as to obtain the probability that the input data belongs to each category, namely the probability that the equipment has no failure and the probability corresponding to each failure category, and taking the category corresponding to the maximum probability value as the category behind the extrusion molding equipment to be detected. For example, when the probability corresponding to no fault of the equipment is the maximum, the extrusion equipment to be detected cannot be subsequently faulted, and when the probability corresponding to a certain equipment fault type D is the maximum, the extrusion equipment to be detected cannot be subsequently faulted, and the fault type is D.

Step S3: and if the extrusion molding equipment to be detected has faults, early warning is carried out according to the fault type.

Through the step S2, whether the extrusion equipment to be detected is in failure or not can be predicted, and if the extrusion equipment to be detected is in failure, early warning is carried out according to the corresponding failure type. In this embodiment, the higher the severity of the fault in the fault category, the higher the warning level, otherwise, the lower the warning level. For example, when the severity of the fault category is high, sound and light early warning is performed, and when the severity of the fault category is low, light warning is performed.

According to the artificial intelligence-based extrusion molding equipment fault early warning method, the fault which is likely to occur to the equipment can be early warned before the fault occurs, the key information of the fault is provided for maintenance personnel, the equipment is prevented from being faulted, the maintenance efficiency is improved, the continuity of the operation of the extrusion molding equipment is ensured, the mean fault interval time is prolonged, and the qualification rate of products is improved. During actual detection, the monitoring video at the current time does not need to be analyzed, only the log text needs to be analyzed, the data dimensionality is reduced, and the time complexity of the early warning method is reduced.

The embodiment also provides an artificial intelligence based extrusion molding equipment fault early warning system, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor is coupled with the memory, and the processor realizes the artificial intelligence based extrusion molding equipment fault early warning method when executing the computer program. Since the method for early warning of the failure of the extrusion molding equipment based on artificial intelligence is described in detail in the above, it is not described herein again.

It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.