PLC channel health state estimation algorithm based on gating neural network model

Abstract

A PLC channel health state estimation algorithm based on a gating neural network model uses the gating neural network model, combines a large number of data sets of PLC parameters with qualified quality, adopts a machine learning method to predict PLC channel data transmission time, and assists in adopting a receiving and sending packet confirmation mode to assess channel transmission quality, thereby analyzing the health state of the PLC channel. And meanwhile, the unqualified PLC channel is further verified, the reason for the unqualified channel is diagnosed and analyzed, and corresponding health value evaluation, defect information and lack-of-sale suggestions are given. According to the method, the PLC channel is detected, the channel quality is fed back in a health value form, the channel quality is subjected to trend analysis, unqualified channels are avoided and prevented, the channels with unqualified health values can be subjected to segmented analysis, and reasonable suggestions are given, so that the detection range of the defect elimination process is narrowed, and manpower and material resources are saved.

Description

Technical Field

[0001] The present invention belongs to the technical field of PLC channel health state prediction, and in particular relates to a PLC channel health state prediction algorithm based on a gated neural network model. Background Art

[0002] In recent years, with the development of society, PLC programmable controllers have been widely used in industrial production. At the same time, the requirements of technicians for their use are also increasing year by year, so the requirements for the normal and stable operation of the system are also getting higher and higher. Programmable controller (PLC) is an electronic device for digital operation designed specifically for use in industrial environments. It uses a programmable memory to store instructions for performing logical operations, sequential operations, timing, counting and arithmetic operations, and can control various types of machinery or production processes through the input and output of switching or analog quantities. However, due to the complex industrial field environment, PLC may encounter channel failure problems during the input and output of data, resulting in abnormal transmission and reception. At the same time, maintaining the health of the PLC transmission channel, shortening the response time and ensuring the transmission accuracy are also one of the focuses of instrument equipment maintenance.

[0003] The current hydropower station monitoring system uses 4-20mA current to transmit data for PLC channel monitoring. The judgment condition for channel failure is that the current value exceeds the upper or lower limit, that is, when the current value sent to the host computer is less than 4mA or greater than 20mA, the host computer displays a channel failure.

[0004] The problems encountered at this stage include the following: 1) Channel faults can only be displayed when the communication path is completely disconnected, and minor damage to the channel cannot be identified; 2) It is impossible to judge the deviation and delay caused by channel aging to communication; 3) Unable to monitor the health status of the PLC and unable to identify the aging of the PLC internal modules and processing delays; 4) It is impossible to identify the specific fault location of a channel fault, and the entire channel needs to be checked in sequence, wasting manpower and material resources. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a PLC channel health status prediction algorithm based on a gated neural network model. The patent of the present invention detects the PLC channel, feeds back the channel quality in the form of a health value, makes a trend analysis of the channel quality, avoids and prevents unqualified channels, and can perform segmented analysis on channels with unqualified health values, and give reasonable suggestions, thereby narrowing the inspection scope of the fault elimination process and saving manpower and material resources.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: A PLC channel health status prediction method based on a gated neural network model, the steps are: 1. Predict the transmission time: Step (1), building a model: first, the PLC response time is used as the target value of the MPNN model for comparison; secondly, the factors that may affect the PLC response time are determined as characteristic values; finally, the main components in the PLC channel are counted to build a topology model; Step (2), clean the data; Step (3), using the splicing point eigenvalues ​​and edge eigenvalues ​​as eigenvectors, iteratively aggregates adjacent node information and updates its own node information multiple times, and through the process of message passing and aggregation, continuously updates the representation of each node and converges into the representation of the entire graph; the entire graph is the aggregation of all directed graphs represented by each PLC branch formed by the constructed model.

[0007] Step (4), use the readout operation to merge the representations of the graph feature vectors of all nodes into a global representation; the readout function is a term in (graph neural network / graph convolutional neural network / graph representation learning / graph classification). In graph representation learning, the model usually obtains node representations of the graph, which are the upstream basis. In downstream applications such as graph classification, we usually need a feature representation of the entire graph, not just a feature representation of all nodes in the graph. At this time, the readout function obtains the feature representation of the entire graph by aggregating node features. Readout can use the summation and average of all nodes in the graph as the output of the graph.

[0008] Step (5), input the obtained vector into the fully connected layer as the output of the time regression task; Step (6), calculate the loss using the mean square error loss function between the output of the fully connected layer and the real time, and use the Adam optimizer for optimization; Step (7), evaluate and modify the model; 2. Calculate the transmission accuracy; 3. Calculate channel health value and predict fault points.

[0009] Preferably, in step (1), the main components include the following: A) Relays and contactors; B) components or equipment for switches, limit positions, safety protection and on-site operation; C) Sub-equipment in the PLC system, including junction boxes, wire terminals, bolts and nuts; D) Sensors and instruments; Based on the different PLC signal processing and transmission times of different components, they are classified and labeled, and components of different models are subdivided.

[0010] Preferably, in step (1), when constructing the model, the factors affecting the signal transmission performance mainly include temperature, surrounding magnetic field strength, transmission medium material and length; therefore, the model is constructed as follows: Each branch of PLC signal transmission is represented as a directed graph G=(V, E), where each node represents a component and the edges between nodes represent the connection relationship between components. The point characteristic values ​​include the component number, the ambient temperature of the control cabinet where the component is located, and whether there is a strong magnetic field around the component. 1 means yes, and 0 means no. The edge eigenvalues ​​include the connection material between components. Preferably, the data cleaning process of step (2) is as follows: collecting PLC channel topology and transmission time of normal quality to construct a data set; PLC adopts NTP time synchronization method, and the transmission time is obtained by subtracting the sending time from the receiving time.

[0011] Preferably, in step (3), the eigenvalues ​​of the splicing points and the eigenvalues ​​of the edges are used as eigenvectors, and formula (2.1), formula (2.2) and formula (2.3) are used to perform multiple iterations to aggregate the information of adjacent nodes and update the information of the own nodes. Through the process of message transmission and aggregation, the representation of each node is continuously updated and aggregated into the representation of the entire graph: (2.1); (2.2); (2.3); Among them, R is the available eigenvalue after iteration, i and j are custom update functions used to process and update node status, and is a custom readout function used to extract the representation of the entire graph; is the eigenvalue of node v after t iterations, is the eigenvalue of its neighboring node w, is the eigenvalue of the edge between two nodes, through its designed The function is aggregated to obtain the updated feature value of its own node v , and then use The function aggregates itself without iteration to obtain new node eigenvalues. Each node aggregates with its two adjacent nodes once, which is called an iteration. It updates its own eigenvalues ​​by interacting with its adjacent nodes and edges. Finally, it uses formula (2.3) to perform vector multiplication between itself and the previous iteration result to obtain the final available eigenvalue. Normalize all edge eigenvalues ​​and scale all features to the same value range. In addition, when using the gradient descent algorithm for optimization, normalization can speed up the convergence of the algorithm and reach the optimal solution faster. Use formula (2.4) for normalization: (2.4); Where x is the original data, is the minimum value of the data, is the maximum value of the data.

[0012] Preferably, in step (4), the readout operation aggregates the representations of all nodes to generate a 128-dimensional global vector; this vector is input into a fully connected layer to output the time prediction result.

[0013] Preferably, step (5) inputs the obtained vector into the fully connected layer as the output of the time regression task: During training, in order to prevent the neural network from overfitting and improve the generalization ability; a Dropout operation is set; Dropout randomly selects some neurons and sets their outputs to 0, so that these neurons are ignored in this round of forward propagation and back propagation; for each layer of the neural network, during the forward propagation, Dropout randomly selects some neurons with probability p and sets their outputs to 0, so that the outputs of these neurons are ignored in the subsequent layers; during the back propagation, Dropout also sets the gradients of the corresponding neurons to 0, so that the contribution of these neurons to the loss function is ignored.

[0014] The dropout is expressed as: when training a deep neural network, we can randomly discard some neurons (and discard their corresponding connections) to avoid overfitting. This method is called dropout. The neurons selected for discarding are random each time. The simplest method is to set a fixed probability p for discarding. When training a neural network model, dropout is often used to make the model more generalizable and prevent overfitting. The essence of dropout is to make some dimensions of the input network data become 0 with a certain probability, which can make the model training more effective.

[0015] Use the activation function Relu of formula (2.5) to retain the features of the input vector and map it to the output end to obtain the output time prediction result: (2.5).

[0016] Preferably, in step (7), model evaluation and modification include modification of hyperparameters, screening of data, and improvement and optimization of models and codes.

[0017] Preferably, the transmission accuracy rate is calculated as follows: the sending PLC randomly sends out several data frames and sets the parity check bit; the receiving PLC uses the parity check method to verify the accuracy of the data transmission and returns a confirmation frame to the sender; the transmission accuracy rate is calculated based on the ratio of the confirmation frame to the sent frame.

[0018] Preferably, the channel health value calculation and fault point prediction method is: for the calculation of the channel health value, the weight of the transmission time deviation and the transmission accuracy ratio are dynamically adjusted by weighted average: (2.6); in, is the channel health value, a and b are weighted coefficients respectively and a+b=1, is the transmission accuracy, t is the predicted transmission time, is the actual transmission time; For PLC channels with a health value lower than 0.6 or a transmission accuracy lower than 0.8, a prompt indicating that the PLC channel is in poor health is given, and the transmission time prediction value label of each channel segment is compared with the actual transmission time. Channel segments with larger errors are marked and priority inspection suggestions are given. If the transmission time is close to the predicted time, it can be preliminarily inferred that the internal module of the PLC is aging and the command processing speed has slowed down.

[0019] The present invention can achieve the following beneficial effects: The patented invention detects PLC channels, feeds back channel quality in the form of health values, and makes a trend analysis of channel quality, which subverts the traditional channel maintenance mode of "problem first, then inspection", avoids and prevents unqualified channels, and can perform segmented analysis on channels with unqualified health values, and give reasonable suggestions, thereby narrowing the inspection scope of the fault elimination process and saving manpower and material resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 This is a diagram of the PLC channel topology model of the present invention; Figure 2 This is a diagram showing the working principle of the activation function of the present invention. DETAILED DESCRIPTION

[0021] The present invention uses a gated neural network model (MPNN) to construct a PLC channel model, and analyzes the model structure, characteristic values ​​and target values ​​of a normal PLC channel in large quantities, thereby comparing them with the target values ​​of the PLC channel to be detected, and predicting a reasonable channel health value. For unhealthy channels, the health value of each branch is analyzed one by one by sending a package, so as to find out the fault point that affects the health of the channel. The specific scheme is as follows: A PLC channel health status estimation method based on a gated neural network model evaluates the health status of the PLC channel by combining transmission time and transmission accuracy, and performs diagnostic verification on channels with unqualified health status to determine the fault point. The steps are: 1. Transmission time prediction algorithm: (1) Construct the model. First, the PLC response time is used as the target value of the MPNN model for comparison; second, the factors that may affect the PLC response time are determined as characteristic values; finally, the main components in the PLC channel are counted to construct the topology model. In the entire PLC control system, the most likely place for failure is on-site, and the most likely place for failure on-site is in the following aspects.

[0022] A) Relays and contactors. In the daily maintenance of PLC control systems, the largest consumption of electrical spare parts is various types of relays or air switches. In addition to the quality of the product itself, it is mainly due to the harsh on-site environment. For example, the contact points of contactors exposed to the production environment are prone to sparking or oxidation, gradually heating up and deforming until they cannot be used.

[0023] B) Some components or equipment on switches, limit positions, safety protection and on-site operations. The reason may be long-term wear and tear, or rust and aging due to long-term disuse.

[0024] C) Sub-equipment in the PLC system, such as junction boxes, wire terminals, bolts and nuts, etc.

[0025] D) Sensors and instruments.

[0026] Based on the different PLC signal processing and transmission times of different components, they are classified and labeled, such as 1-PLC, 2-relay, 3-contactor, 4-voltmeter... For the sake of accuracy, different types of components can also be subdivided.

[0027] The factors that affect signal transmission performance mainly include temperature, surrounding magnetic field strength, transmission medium material and length, so the model is constructed as follows: each branch of PLC signal transmission is represented as a directed graph G=(V, E), each node represents a component, and the edges between nodes represent the connection relationship between components. The point eigenvalues ​​include the component number, the ambient temperature of the control cabinet where the component is located, and whether there is a strong magnetic field around the component, 1 for yes, 0 for no; the edge eigenvalues ​​include the connection material between components, such as 1-optical fiber, 2-copper wire, 3-aluminum wire... and the material length. The model diagram is as follows Figure 1 shown.

[0028] (2) Clean the data, collect the PLC channel topology and transmission time of normal quality, and construct a data set. PLC uses NTP time synchronization, and the transmission time can be obtained by subtracting the sending time from the receiving time.

[0029] (3) The eigenvalues ​​of the splicing points and the edge eigenvalues ​​are used as eigenvectors, and formulas (2.1), (2.2), and (2.3) are used to perform multiple iterations to aggregate the information of adjacent nodes and update the information of the own nodes. Through the process of message passing and aggregation, the representation of each node is continuously updated and converged into the representation of the entire graph.

[0030] (2.1); (2.2); (2.3); in, is the eigenvalue of node v after t iterations, is the eigenvalue of its neighboring node w, is the eigenvalue of the edge between two nodes, through its designed The function is aggregated to obtain the updated feature value of its own node v , and then use The function aggregates itself without iteration to obtain new node eigenvalues. Each node aggregates with its two adjacent nodes once, which is called an iteration. This is also the characteristic of message passing neural network: updating its own eigenvalues ​​by interacting with its adjacent nodes and edges. Finally, use formula (2.3) to perform vector multiplication between itself and the previous iteration result to obtain the final available eigenvalue.

[0031] When reading in the path, the point feature value is only 0 or 1 or tens, while the edge feature value may reach hundreds, which will cause the model to be more affected by features with a larger value range, and for features with a smaller value range, the model weight will become very small. To solve this problem, we normalize all edge feature values. We can scale all features to the same value range to avoid this problem. In addition, when using the gradient descent algorithm for optimization, normalization can speed up the convergence of the algorithm and reach the optimal solution faster. Use formula (2.4) for normalization: (2.4); Where x is the original data, is the minimum value of the data, is the maximum value of the data.

[0032] (4) Use the readout operation to merge the representations of the graph feature vectors of all nodes into a global representation. In this model, the readout operation aggregates the representations of all nodes to generate a 128-dimensional global vector. This vector is input into a fully connected layer to output the time prediction result.

[0033] (5) The obtained vector is input into the fully connected layer as the output of the time regression task; during training, in order to prevent the neural network from overfitting and improve the generalization ability. A Dropout operation is set. Dropout randomly selects some neurons and sets their outputs to 0, so that these neurons are ignored in this round of forward and back propagation. Specifically, for each layer of the neural network, during the forward pass, Dropout will randomly select some neurons with probability p and set their outputs to 0, so that the outputs of these neurons are ignored in subsequent layers. During back propagation, Dropout will also set the gradients of the corresponding neurons to 0, so that the contribution of these neurons to the loss function is ignored.

[0034] By randomly discarding the output of some neurons, Dropout can reduce the co-adaptation phenomenon in the neural network, that is, the mutual dependence between neurons, and reduce the complexity of the neural network. The role of Dropout is similar to ensemble learning, which can make the neural network more robust and able to adapt to new data sets, thereby improving the generalization ability of the model.

[0035] Use the activation function Relu in formula (2.5) to retain the features of the input vector and map it to the output to obtain the output time prediction result.

[0036] (2.5); like Figure 2 As shown in the figure, it depicts the working principle of the activation function Relu. Each neuron node in the neural network accepts the output value of the neuron in the previous layer as the input value of the neuron, and passes the input value to the next layer. The neuron node in the input layer will directly pass the input attribute value to the next layer. In a multi-layer neural network, there is a functional relationship between the output of the upper layer node and the input of the lower layer node. This function is called the activation function. Simply put, the activation function retains and maps the features through a function, that is, it is responsible for mapping the input of the neuron to the output. In this experiment, the activation function is responsible for converting the vector obtained in step (5) into the output of the fully connected layer, that is, the prediction time.

[0037] (6) The output of the fully connected layer is compared with the real time using the mean square error loss function to calculate the loss, and the Adam optimizer is used for optimization. During the training process, the parameters of the neural network are updated through the back propagation algorithm to gradually improve the prediction accuracy of the model. (7) Evaluate and modify the model, including modification of hyperparameters, data screening, and improvement and optimization of the model and code.

[0038] 2. Transmission accuracy calculation: The sending PLC randomly sends out several data frames and sets the parity bit. The receiving PLC verifies the accuracy of data transmission by using parity verification and returns a confirmation frame to the sender. The transmission accuracy is calculated based on the ratio of confirmation frames to sent frames.

[0039] 3. Channel health value calculation and fault point prediction: For the calculation of channel health value, the weighted average method can be used to dynamically adjust the weight of transmission time deviation and transmission accuracy ratio.

[0040] (2.6); in, is the channel health value, a and b are weighted coefficients respectively and a+b=1, is the transmission accuracy, t is the predicted transmission time, is the actual transmission time.

[0041] For PLC channels with health values ​​lower than 0.6 or transmission accuracy lower than 0.8, a poor health status prompt is given, and the transmission time prediction value label of each channel is compared with the actual transmission time, and the channel segments with large errors are marked and priority inspection suggestions are given. If the transmission time is close to the predicted time, it can be preliminarily inferred that the internal module of the PLC is aging and the command processing speed is slow.

[0042] In this embodiment, the experimental tools and parameter settings are shown in Table 1, Table 2, and Table 3: Table 5.1 Development Tools name Version Number illustrate Python 3.9.0 programming language Pycharm 2023.1 Python Coding IDE Table 5.2 Python libraries name Version Number illustrate Pytorch 1.8.1 Machine Learning Frameworks scikit-learn 1.0.2 Machine Learning Frameworks Pandas 1.1.5 Dataset reading numpy 1.18.5 Dataset preprocessing joblib 1.2.0 Save training log Deep Graph Library 0.7.0 Graph Neural Network Algorithms Table 5.3 Training hyperparameters Hyperparameters Numeric Epoch 1000 Batch_Size 50 Learning_Rate 0.001 hidden_state_size 15 The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A PLC channel health status prediction algorithm based on a gated neural network model, characterized in that: To make a transmission time prediction: Step (1), building a model: first, the PLC response time is used as the target value of the MPNN model for comparison; secondly, the factors that may affect the PLC response time are determined as characteristic values; finally, the main components in the PLC channel are counted to build a topology model; Step (2), clean the data; Step (3): Use the feature values ​​of the splicing points and the edge feature values ​​as feature vectors, perform multiple iterations to aggregate the information of adjacent nodes and update the node information itself. Through the process of message passing and aggregation, the representation of each node is continuously updated and aggregated into the representation of the entire graph. Step (4), use the readout operation to merge the representations of the graph feature vectors of all nodes into a global representation; Step (5), input the obtained vector into the fully connected layer as the output of the time regression task; Step (6), calculate the loss using the mean square error loss function between the output of the fully connected layer and the real time, and use the Adam optimizer for optimization; Step (7), evaluate and modify the model; Calculate the transmission accuracy; Calculate channel health values ​​and predict fault points.

2. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: In step (1), the main components include the following: A) Relays and contactors; B) components or equipment for switches, limit positions, safety protection and on-site operation; C) Sub-equipment in the PLC system, including junction boxes, wire terminals, bolts and nuts; D) Sensors and instruments; Based on the different PLC signal processing and transmission times of different components, they are classified and labeled, and components of different models are subdivided.

3. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 2 is characterized in that: In step (1), when constructing the model, the factors that affect the signal transmission performance mainly include temperature, surrounding magnetic field strength, transmission medium material and length; therefore, the model is constructed as follows: Each branch of PLC signal transmission is represented as a directed graph G=(V, E), where each node represents a component and the edges between nodes represent the connection relationship between components. The point characteristic values ​​include the component number, the ambient temperature of the control cabinet where the component is located, and whether there is a strong magnetic field around the component. 1 means yes, and 0 means no. The edge eigenvalues ​​include the connection material between components.

4. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: The data cleaning process in step (2) is as follows: collect the PLC channel topology and transmission time of normal quality and construct a data set; PLC uses NTP time synchronization, and the transmission time is obtained by subtracting the sending time from the receiving time.

5. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: In step (3), the eigenvalues ​​of the splicing points and the edge eigenvalues ​​are used as eigenvectors, and formula (2.1), formula (2.2) and formula (2.3) are used to perform multiple iterations to aggregate the information of adjacent nodes and update the information of the own nodes. Through the process of message passing and aggregation, the representation of each node is continuously updated and converged into the representation of the entire graph: （2.1）； （2.2）； （2.3）； Among them, R is the available eigenvalue after iteration, i and j are custom update functions used to process and update node status, and is a custom readout function used to extract the representation of the entire graph; is the eigenvalue of node v after t iterations, is the eigenvalue of its neighboring node w, is the eigenvalue of the edge between two nodes, through its designed The function is aggregated to obtain the updated feature value of its own node v , and then use The function aggregates itself without iteration to obtain new node eigenvalues. Each node aggregates with its two adjacent nodes once, which is called an iteration. It updates its own eigenvalues ​​by interacting with its adjacent nodes and edges. Finally, it uses formula (2.3) to perform vector multiplication between itself and the previous iteration result to obtain the final available eigenvalue. Normalize all edge eigenvalues ​​and scale all features to the same value range. In addition, when using the gradient descent algorithm for optimization, normalization can speed up the convergence of the algorithm and reach the optimal solution faster. Use formula (2.4) for normalization: （2.4）； Where x is the original data, is the minimum value of the data, is the maximum value of the data.

6. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: In step (4), the readout operation aggregates the representations of all nodes to generate a 128-dimensional global vector; this vector is input into a fully connected layer to output the time prediction result.

7. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: Step (5) Input the obtained vector into the fully connected layer as the output of the time regression task: During training, in order to prevent the neural network from overfitting and improve the generalization ability; Dropout operation is set; Dropout will randomly select some neurons and set their outputs to 0, so that these neurons are ignored in this round of forward propagation and back propagation; For each layer of the neural network, during the forward pass, Dropout will randomly select some neurons with probability p and set their outputs to 0, so that the outputs of these neurons are ignored in the subsequent layers; During back propagation, Dropout will also set the gradients of the corresponding neurons to 0, so that the contribution of these neurons to the loss function is ignored; Use the activation function Relu of formula (2.5) to retain the features of the input vector and map it to the output end to obtain the output time prediction result: （2.5）。 8. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: In step (7), the model is evaluated and modified, including modification of hyperparameters, screening of data, and improvement and optimization of the model and code.

9. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: The transmission accuracy rate is calculated as follows: the sending PLC randomly sends out several data frames and sets the parity check bit. The receiving PLC uses the parity check method to verify the accuracy of the data transmission and returns a confirmation frame to the sender. The transmission accuracy rate is calculated based on the ratio of the confirmation frame to the sent frame.

10. The PLC channel health status prediction algorithm based on the gated neural network model according to claim 1 is characterized in that: The channel health value calculation and fault point prediction method is as follows: For the calculation of the channel health value, the weighted average method is used to dynamically adjust the transmission time deviation and the transmission accuracy ratio weight: (2.6)； in, is the channel health value, a and b are weighted coefficients respectively and a+b=1, is the transmission accuracy, t is the predicted transmission time, is the actual transmission time; For PLC channels with a health value lower than 0.6 or a transmission accuracy lower than 0.8, a prompt indicating that the PLC channel is in poor health is given, and the transmission time prediction value label of each channel segment is compared with the actual transmission time. Channel segments with larger errors are marked and priority inspection suggestions are given. If the transmission time is close to the predicted time, it can be preliminarily inferred that the internal module of the PLC is aging and the command processing speed has slowed down.

Images