Equipment fault detection data analysis method and system for digital factory

Abstract

The invention discloses an equipment fault detection data analysis method and system for a digital factory. The method comprises the following steps: S1, synchronously acquiring comprehensive maintenance data of equipment by using a high-precision clock synchronization technology; s2, constructing a graph neural network by taking the position data of the devices as graph nodes and taking function association and data interaction data between the devices as edges of a graph, constructing a device association model based on the graph neural network, and taking the state data of the devices as input and output state association factors of the device association model; s3, taking the comprehensive maintenance data and the state correlation factor as state input of a reinforcement learning agent, training the agent by adopting a Q learning algorithm, inputting real-time state data into the agent, outputting a state score, setting an abnormal state threshold value, when the state score is lower than the abnormal state threshold value, marking the corresponding real-time data as abnormal, and when the state score is lower than the abnormal state threshold value, marking the corresponding real-time data as abnormal data. According to the method, unnecessary shutdown and maintenance can be reduced under the condition that the operation state of the equipment is complex and changeable.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method and system for analyzing equipment fault detection data for a digital factory. Background Art

[0002] As digital factories are developing rapidly, the stability and reliability of equipment operation have become key factors in determining production efficiency and product quality. In digital factories, there are many types of equipment with different functions, and the equipment is interconnected and works together, forming a complex production system.

[0003] At present, traditional equipment fault data analysis methods in digital factories mostly rely on regular maintenance, simple threshold alarms or analysis based on a single data type. These methods cannot fully capture the complex operating status of the equipment and are prone to missed detection and false detection of faults. For example, in some factories, for fault detection of motor equipment, only a simple threshold is set based on the current value, and an alarm is issued when the current exceeds the threshold. However, this single data type analysis method cannot fully capture the complex operating status of the equipment. According to statistics, in factories using such traditional methods, the missed detection rate of equipment faults is as high as 25%, and the false detection rate is as high as 20%. This not only leads to frequent sudden shutdowns in the production process, but also causes unnecessary waste of maintenance resources. Therefore, a method and system for equipment fault detection data analysis for digital factories are proposed. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention proposes the following technical solutions: A method for analyzing equipment fault detection data for a digital factory, the method steps comprising: S1: Use high-precision clock synchronization technology to synchronize and collect comprehensive maintenance data of equipment; S2: The location data of the devices is used as graph nodes, and the functional associations and data interaction data between devices are used as graph edges to construct a graph neural network. A device association model is built based on the graph neural network. The status data of the devices is used as the input of the device association model, and the status association factor is output. S3: The comprehensive maintenance data and the state correlation factor are used as the state input of the reinforcement learning agent, and the Q-learning algorithm is used to train the agent to obtain the reinforcement learning agent. The real-time state data is input into the reinforcement learning agent, the state score is output, and the abnormal state threshold is set. When the state score is lower than the abnormal state threshold, the corresponding real-time data is marked as abnormal.

[0005] The comprehensive maintenance data includes maintenance time, maintenance type, specified maintenance cost information and maintenance behavior data.

[0006] The specified maintenance cost information includes parts cost data, labor cost data and special cost data; The process of obtaining the special cost data is as follows: The pressure sensor measures the pressure at a fixed sampling frequency to obtain pressure data ; The wear sensor periodically detects the wear of the tool and obtains the wear value. ; Use blockchain technology to record the full life cycle information of tools ; The pressure data , wear value and tool life cycle information Integrate as input data for the cost relationship model and establish a cost relationship model to obtain special cost data .

[0007] The construction process of the cost relationship model is as follows: Based on pressure data , wear value and tool life cycle information Construct a new feature combination, expressed as ,in, is the collection time interval, i is the index; Through the random forest regression model, the feature combination is used as the input variable and the actual loss cost of the tool is used as the output variable to obtain the cost relationship model.

[0008] The process of acquiring the location data, function-related data and interaction data of the device is as follows: The process of obtaining location data is as follows: Equip each device in the digital factory with a UWB tag and obtain the 3D coordinates of the device as location data , the location data of each device is represented as a node feature vector ,in Indicates the UWB tag number; The process of acquiring functional association data is as follows: Extract the functional relationship between devices from the digital factory operation log and use numerical Indicates the weight of the functional association between devices, and the value range is , 0 means no association, Indicates strong association, and organizes the functional association weights between all devices into a functional association adjacency matrix , i.e., functional association data; The interactive data acquisition process is: Deploy data monitoring software at the data transmission interface of the device to monitor the frequency and amount of data transmission, count the number of times and amount of data transmitted from device to device per unit time, and obtain the data interaction weight after normalization. , and organized into an interactive adjacency matrix , that is, interaction data.

[0009] The process of constructing the graph neural network is as follows: The location data is used as the nodes of the graph, and the functional associations and data interaction data between devices are used as the edges of the graph to obtain the graph; Define the graph as ,in, is a node set consisting of nodes corresponding to device location data. is an edge set consisting of functional associations and data interaction data between devices. The edge weight matrix is ​​the weighted sum of the adjacency matrices of functional associations and data interaction data, expressed as ,in, Represents the functional association adjacency matrix The weight of Represents the interaction adjacency matrix The weight of Select graph convolutional network as the basic architecture of graph neural network model, and define message passing and feature update mechanism to update its own features by aggregating information of neighboring nodes; Determine the training target as the state correlation factor between devices; The network is trained by dividing the collected device-related data into training set, validation set and test set; After training, we obtain the graph neural network.

[0010] The device association model construction process is as follows: After preprocessing, the device status data is fused with the graph neural network as the input of the graph neural network; Using historical equipment failure data as supervision information, define the loss function Train the device association model and minimize the loss function to complete the model training.

[0011] The construction process of the reinforcement learning agent is as follows: Reinforcement learning agents include state space, action space, and reward function; Composing state space based on comprehensive maintenance data and state correlation factors; The action space is defined as the judgment action of the device state. Indicates that the real-time data of the device is normal and the action =1 means that the real-time data of the device is judged to be abnormal; Define a reward function. When the agent correctly judges the device status, it will be given a positive reward. When it makes an incorrect judgment, it will be given a negative reward. Set the device actual status to , Indicates normal, Indicates abnormality; Then the agent is trained through the Q learning algorithm to obtain the reinforcement learning agent.

[0012] The process of training an agent through the Q learning algorithm to obtain a reinforcement learning agent is as follows: Determine the agent's state, action space, and reward function, initialize the Q network parameters and experience replay buffer, and set the total training time step and discount factor; At each time step, observe the state of the agent, through The greedy strategy selects an action, obtains the reward and new state of the environment feedback after execution, and stores them in the experience replay buffer; Randomly extract samples from the experience replay buffer and obtain the Q value of the target, calculate it at the same time, and then update the Q network parameters through the back propagation algorithm; The loss gradually decreases during training, and the Q network training steps are repeated to reach the total training time step until the Q learning network converges and the training is completed.

[0013] An equipment fault detection data analysis system for a digital factory, comprising: Synchronous acquisition module: uses high-precision clock synchronization technology to synchronously acquire comprehensive maintenance data of equipment; State association acquisition module: The location data of the device is used as the graph node, and the functional association and data interaction data between the devices are used as the edges of the graph to build a graph neural network. The device association model is built based on the graph neural network, and the state data of the device is used as the input of the device association model to output the state association factor. Comprehensive fault detection module: The comprehensive maintenance data and state correlation factors are used as the state input of the reinforcement learning agent, and the Q learning algorithm is used to train the agent to obtain the reinforcement learning agent. The real-time state data is input into the reinforcement learning agent, the state score is output, and the abnormal state threshold is set. , when the state score is lower than the abnormal state threshold , the corresponding real-time data is marked as abnormal.

[0014] The present invention has the following beneficial effects: In the present invention, firstly, comprehensive maintenance data is collected by adopting high-precision clock synchronization technology, covering maintenance time, type, cost and behavior data. Special cost data is obtained by using the Internet of Things and blockchain technology, and maintenance behavior data is integrated with sensors through handheld terminals. These multi-dimensional data provide a more comprehensive basis for fault judgment, avoiding missed detection and false detection due to single data; Secondly, we build a graph neural network and equipment association model with equipment location as nodes, functional association and data interaction as edges, output state association factors, and deeply explore the complex relationship between equipment. In a complex digital factory, a device failure may trigger a chain reaction. Through this model, we can quickly locate the source of the failure and the affected equipment to avoid the expansion of the scope of the failure. Finally, the comprehensive maintenance data and status correlation factors are used as the input of the reinforcement learning agent, which is trained using the Q learning algorithm. The status score is output and compared with the abnormal threshold to determine the equipment status. This method can dynamically adapt to changes in equipment operation. Compared with traditional fixed threshold alarms, it can detect faults more flexibly and accurately and reduce the rate of missed detection and false detection. For situations where the equipment operation status is complex and changeable, the agent can accurately judge faults based on real-time data and reduce unnecessary downtime and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A method step diagram of a method and system for analyzing equipment fault detection data for a digital factory proposed by the present invention; Figure 2 This is a system block diagram of a method and system for analyzing equipment fault detection data for a digital factory proposed by the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Embodiment 1: Figure 1 As shown, the present invention proposes a method for analyzing equipment fault detection data for a digital factory, comprising: S1: using high-precision clock synchronization technology to synchronously collect comprehensive maintenance data of equipment; Use high-precision clock synchronization technology to extract comprehensive maintenance data from the equipment maintenance process of the digital factory; Comprehensive maintenance data includes maintenance time, maintenance type (preventive maintenance, corrective maintenance), specified maintenance cost information, and maintenance behavior data; The process of obtaining the specified maintenance cost information is as follows: Maintenance costs include parts cost data, labor cost data and special cost data; The process of obtaining special cost data is as follows: For special costs (tool loss) used in the maintenance process, the blockchain technology is deployed and applied through IoT devices to obtain them; Deployment through IoT devices: Install micro IoT sensors on the tools used in digital factory equipment, including pressure sensors and wear sensors; The pressure sensor measures the pressure at a fixed sampling frequency, and obtains a pressure data each time it is sampled. , and record the sampling time at the same time, expressed as ,in is the frequency of pressure data collected; The wear sensor periodically detects the wear of the tool and obtains a value representing the amount of wear each time it is detected. , and record the corresponding time, expressed as ,in, is the number of cycles of wear data collected; Specifically, the pressure sensor can monitor the pressure and duration of the tool during use in real time, and the wear sensor uses optical or electromagnetic induction principles to monitor the wear of the tool in real time; By applying blockchain technology: The whole life cycle information of the tool is recorded by blockchain technology, including relevant operation information of the tool (user, time of use, and place of use), which is recorded on the distributed ledger in the form of blockchain smart contracts. At the same time, the real-time status data of the tool transmitted by the IoT platform will also be packaged into a transaction record of the blockchain, expressed as ,in, Represents the number of data points over the entire life cycle, , Indicates the user, Indicates the usage time, Indicates the place of use; Based on pressure data , wear value and tool life cycle information Integrate as input data for the cost relationship model and establish a cost relationship model for tool loss costs to obtain special cost data ; The construction process of the cost relationship model is as follows: Based on pressure data , wear value and tool life cycle information Construct a new feature combination, expressed as ,in, is the collection time interval, i is the index; Through the random forest regression model, the feature combination is used as the input variable, and the actual loss cost of the tool is used as the output variable (which can be determined by historical tool loss records and cost accounting data). The training set and test set are divided according to the ratio of 70%-30%. The random forest regression model is trained using the training set data, and the model's hyperparameters (the number of decision trees and the maximum depth) are adjusted; The trained model is evaluated using the test set data, and the mean square error (MSE) is used as the evaluation indicator. The formula is expressed as:

[0018] in, is the number of test set samples, is the actual tool loss cost, Tool wear and tear costs predicted by the model; After the evaluation indicators are completed, a trained cost relationship model is obtained. Input the real-time tool usage data into the trained cost relationship model, and the model outputs the predicted special cost data (tool loss) ; The process of obtaining maintenance behavior data is as follows: Use a handheld terminal to scan the QR code or RFID tag on the device to obtain the unique identification of the device Then record the tools used and the parts replaced, store this information in the form of structured data, collect sensor data, and associate the data collected by the sensor with a tag recorded by the handheld terminal. According to the equipment identification and maintenance information, the data from different sources are integrated to The maintenance behavior data is represented as a multi-tuple , where the multi-tuple Zhong Youzai The measured value of the sensor at time Maintenance operation content, tool information, replaced parts information and unique identification are recorded at all times composition.

[0019] S2: The location data of the devices is used as graph nodes, and the functional associations and data interaction data between devices are used as graph edges to construct a graph neural network. A device association model is built based on the graph neural network. The status data of the devices is used as the input of the device association model, and the status association factor is output. The graph neural network construction process is as follows: Equip each device in the digital factory with a UWB tag to obtain the 3D coordinates of the device as location data in real time , the location data of each device is represented as a node feature vector ,in Indicates the UWB tag number; Extract the functional relationship between devices from the digital factory operation log. For example, if device 1 provides processing raw materials for device 2, then there is a functional relationship between device 1 and device 2. Indicates the weight of the functional association between devices, and the value range is , 0 means no association, Indicates strong association, and organizes the functional association weights between all devices into a functional association adjacency matrix , this function associates the adjacency matrix That is, functional association data; Deploy data monitoring software at the data transmission interface of the device to monitor the data interaction between devices in real time, including the frequency and amount of data transmission, and count the number of times and amount of data transmitted from device to device per unit time. After normalization, the data interaction weight is obtained. , and organized into an interactive adjacency matrix , the interaction adjacency matrix That is, interaction data; With location data as the nodes of the graph, and functional associations and data interaction data between devices as the edges of the graph, the graph is defined as , among which, among which, is a node set consisting of nodes corresponding to device location data. is an edge set, which consists of functional associations and data interaction data between devices. The edge weight matrix is ​​the weighted sum of the weight matrices of functional associations and data interaction data, expressed as ,in, Represents the functional association adjacency matrix The weight of Represents the interaction adjacency matrix The weight of Select graph convolutional network as the basic architecture of graph neural network model, and define message passing and feature update mechanism, that is, nodes update their own features by aggregating information of neighboring nodes. , which is in Layer characteristics The update formula is:

[0020] in, is the layer index, is the normalization constant, For the The characteristics of the layer, For the The weight of the layer, For the The bias term of the layer, For Node The number of neighbor nodes; Determine the training target. During the training process, the target is determined as the state correlation factor between the devices with accurate output; The cross entropy loss function is used for training, and the collected device-related data is divided into a training set, a validation set, and a test set, with a ratio of 70% as a training set, 15% as a validation set, and 10% - 15% as a test set; Graph neural network training process: Use the training set data to train the graph neural network, calculate the gradient of the loss function to the model parameters (weight matrix and bias vector) through the back-propagation algorithm, and then use the optimizer to update the parameters to minimize the loss function. The model performance is regularly evaluated on the validation set, and the hyperparameters are adjusted according to the evaluation results. When the performance of the model on the validation set no longer improves (the loss function no longer decreases), the model is considered to have converged, and the training is stopped. The graph neural network is constructed based on the device location data as the nodes of the graph, and the functional associations and data interaction data between devices as the edges of the graph; The process of building a device association model is as follows: After preprocessing, the device status data is fused with the graph neural network as the input of the graph neural network. Suppose the device status data is , is the number of data points; Specifically, the operating parameters collected by sensor data are state data, and the specific operating parameters include electrical parameters and mechanical parameters. The electrical parameters include voltage, current, power, and resistance, and the mechanical parameters include temperature, vibration, speed, and pressure. Using historical equipment failure data as supervision information, define the loss function Train the device association model and minimize the loss function to complete the model training; The device association model outputs the state association factor between devices ; Specifically, within the device association model, the graph neural network performs calculations based on the pre-trained parameters and the defined message passing and feature update mechanisms. The graph neural network calculates the functional association weight matrix between devices. (Reflects the functional dependence and collaboration between devices. For example, if device A provides raw materials for device B, there is a functional relationship between the two), data interaction weight matrix (Reflects the frequency and amount of data transmission between devices) and device location information (represented by three-dimensional coordinates obtained by the UWB tag equipped with the device), the input device status data is processed, and the node aggregates the information of neighboring nodes (according to specific aggregation rules, such as the comprehensive consideration of neighboring node features, edge weights, and normalization constants in the graph convolutional network), continuously updates its own features, and finally obtains the result that can reflect the degree of status correlation between devices after multiple layers of calculation and feature transformation; S3: The comprehensive maintenance data and the state correlation factor are used as the state input of the reinforcement learning agent, and the Q learning algorithm is used to train the agent to obtain the reinforcement learning agent. The real-time state data is input into the reinforcement learning agent, the state score is output, and the abnormal state threshold is set. When the state score is lower than the abnormal state threshold, the corresponding real-time data is marked as abnormal. The construction process of the reinforcement learning agent is: Reinforcement learning agents include state space, action space, and reward function; Constructing state space: The state space consists of comprehensive maintenance data and state association factors. The comprehensive maintenance data is represented as a coding vector, which is expressed as ,in is the maintenance time vector, is the encoding vector maintaining the type, is the maintenance cost vector, is to maintain behavioral data; Then the state space of the agent is ; Specifically, the state space covers comprehensive maintenance data (such as maintenance time, maintenance type, maintenance cost, etc.) and equipment state correlation factors, which can reflect the operating status of equipment in the digital factory; Constructing action space: The action space is defined as the judgment action of the device status, that is, marking whether the real-time data of the device is abnormal. It can be represented by discrete actions. Define a discrete action ,action Indicates that the real-time data of the device is normal and the action =1 means that the real-time data of the device is judged to be abnormal; Specifically, the action space is the agent’s judgment action on the device status (judgment of normal or abnormal); Define the reward function: When the agent correctly judges the device status (that is, the judgment result is consistent with the actual situation), it will be given a positive reward, and when the judgment is wrong, it will be given a negative reward; Set the device actual status to , Indicates normal, Indicates an abnormality, the agent's judgment action is , the reward function can be defined as:

[0021] The process of training an agent through the Q-learning algorithm is: Determine the agent's state, action space, and reward function, initialize the Q network parameters and experience replay buffer, and set the total training time step and discount factor; At each time step, observe the state of the agent, through The greedy strategy selects an action, obtains the reward and new state of the environment feedback after execution, and stores them in the experience replay buffer; Randomly extract samples from the experience replay buffer and obtain the Q value of the target, while calculating the loss (number of samples), and then update the Q network parameters through the back-propagation algorithm; The loss (number of samples) gradually decreases during training, and the Q network training steps are repeated to reach the total training time step until the Q learning network converges, and finally an intelligent agent trained by the Q learning network is obtained; Specifically, the Q learning network is a neural network used to estimate the value of state-action pairs. The initial values ​​of its parameters will affect the starting point of the agent's learning. At the same time, an experience replay buffer is created to store the experience data (including information such as state, action, reward, and new state) generated by the interaction between the agent and the environment. This helps to break the correlation between data and improve the training effect. In addition, the total training time step and discount factor are set. The total training time step determines the training time of the agent. The discount factor (value range is [0, 1]) is used to measure the importance of future rewards. The closer it is to 1, the more the agent pays attention to long-term rewards. Determine an abnormal state threshold , input the real-time device data into the trained reinforcement learning agent, and the agent outputs the state score ,when When , the corresponding real-time device data is abnormal; Specifically, the abnormal state threshold The method of determining is: In a digital factory, the entire life cycle data of the equipment from its commissioning to the current stage is continuously collected, including the equipment's operating time, cumulative workload, maintenance records, and operating parameters, and a performance degradation curve is constructed based on the entire life cycle data; For example: with the operation time as the horizontal axis and the performance index as the vertical axis, the initial operation time is 0-500 hours, the average processing accuracy deviation is 0.05mm, and the tool wear is 0.1mm. After 1000 hours, the processing accuracy deviation is 0.08mm, and the tool wear is 0.2mm. After 2000 hours, the processing accuracy deviation reaches 0.12mm, and the tool wear is 0.4mm. Connect the data points to form a recession curve. According to the reaction efficiency decay curve, the decay efficiency value of the abnormal decline of the equipment performance index in a specific time period is obtained. That is, in a specific time period of equipment use, when the decline of the equipment performance index (reaction efficiency) exceeds the decline of the performance index in the time period under normal circumstances set in advance or obtained through statistical analysis, the corresponding decay efficiency value is the abnormal state threshold .

[0022] Embodiment 2: Figure 2 As shown, a device fault detection data analysis system for a digital factory includes: Synchronous acquisition module: uses high-precision clock synchronization technology to synchronously acquire comprehensive maintenance data of equipment; State association acquisition module: The location data of the device is used as the graph node, and the functional association and data interaction data between the devices are used as the edges of the graph to build a graph neural network. The device association model is built based on the graph neural network, and the state data of the device is used as the input of the device association model to output the state association factor. Comprehensive fault detection module: The comprehensive maintenance data and state correlation factors are used as the state input of the reinforcement learning agent, and the Q learning algorithm is used to train the agent to obtain the reinforcement learning agent. The real-time state data is input into the reinforcement learning agent, the state score is output, and the abnormal state threshold is set. When the state score is lower than the abnormal state threshold, the corresponding real-time data is marked as abnormal.

[0023] In the application, several formulas involved are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent actual situation. Some coefficients or weights in the formulas are set by technicians in this field according to actual conditions, so they will not be elaborated here.

[0024] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0025] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing equipment fault detection data for a digital factory, characterized in that: The method steps include: S1: Use high-precision clock synchronization technology to synchronize and collect comprehensive maintenance data of equipment; S2: The location data of the devices is used as graph nodes, and the functional associations and data interaction data between devices are used as graph edges to construct a graph neural network. A device association model is built based on the graph neural network. The status data of the devices is used as the input of the device association model, and the status association factor is output. S3: The comprehensive maintenance data and the state correlation factor are used as the state input of the reinforcement learning agent, and the Q-learning algorithm is used to train the agent to obtain the reinforcement learning agent. The real-time state data is input into the reinforcement learning agent, the state score is output, and the abnormal state threshold is set. When the state score is lower than the abnormal state threshold, the corresponding real-time data is marked as abnormal.

2. The method for analyzing equipment fault detection data for a digital factory according to claim 1, characterized in that: The comprehensive maintenance data includes maintenance time, maintenance type, specified maintenance cost information and maintenance behavior data.

3. The equipment fault detection data analysis method for a digital factory according to claim 2, characterized in that: The specified maintenance cost information includes parts cost data, labor cost data and special cost data; The process of obtaining the special cost data is as follows: The pressure sensor measures the pressure at a fixed sampling frequency to obtain pressure data ; The wear sensor periodically detects the wear of the tool and obtains the wear value. ; Use blockchain technology to record the full life cycle information of tools ; The pressure data , wear value and tool life cycle information Integrate as input data for the cost relationship model and establish a cost relationship model to obtain special cost data .

4. The method for analyzing equipment fault detection data for a digital factory according to claim 3, characterized in that: The construction process of the cost relationship model is as follows: Based on pressure data , wear value and tool life cycle information Construct a new feature combination, expressed as ,in, is the collection time interval, i is the index; Through the random forest regression model, the feature combination is used as the input variable and the actual loss cost of the tool is used as the output variable to obtain the cost relationship model.

5. The method for analyzing equipment fault detection data for a digital factory according to claim 1, characterized in that: The process of acquiring the location data, function-related data and interaction data of the device is as follows: The process of obtaining location data is as follows: Equip each device in the digital factory with a UWB tag and obtain the 3D coordinates of the device as location data , the location data of each device is represented as a node feature vector ,in Indicates the UWB tag number; The process of acquiring functional association data is as follows: Extract the functional relationship between devices from the digital factory operation log and use numerical Indicates the weight of the functional association between devices, and the value range is , 0 means no association, Indicates strong association, and organizes the functional association weights between all devices into an adjacency matrix , obtain functional association data; The interactive data acquisition process is: Deploy data monitoring software at the data transmission interface of the device to monitor the frequency and amount of data transmission, count the number of times and amount of data transmitted from device to device per unit time, and obtain the data interaction weight after normalization. , and organized into an adjacency matrix , get the interaction data.

6. The method for analyzing equipment fault detection data for a digital factory according to claim 1, characterized in that: The process of constructing the graph neural network is as follows: The location data is used as the nodes of the graph, and the functional associations and data interaction data between devices are used as the edges of the graph to obtain the graph; Define the graph as ,in, is a node set consisting of nodes corresponding to device location data. is an edge set, which consists of functional associations and data interaction data between devices. The edge weight matrix is ​​the weighted sum of the weight matrices of functional associations and data interaction data, expressed as ,in, Represents the functional association adjacency matrix The weight of Represents the interaction adjacency matrix The weight of Select graph convolutional network as the basic architecture of graph neural network model, and define message passing and feature update mechanism to update its own features by aggregating information of neighboring nodes; Determine the training target as the state correlation factor between devices; The network is trained by dividing the collected device-related data into training set, validation set and test set; After training, we obtain the graph neural network.

7. The method for analyzing equipment fault detection data for a digital factory according to claim 1, characterized in that: The device association model construction process is as follows: After preprocessing, the device status data is fused with the graph neural network as the input of the graph neural network; Using historical equipment failure data as supervision information, define the loss function Train the device association model and minimize the loss function to complete the model training.

8. The method for analyzing equipment fault detection data for a digital factory according to claim 1, characterized in that: The construction process of the reinforcement learning agent is as follows: Reinforcement learning agents include state space, action space, and reward function; Composing state space based on comprehensive maintenance data and state correlation factors; The action space is defined as the judgment action of the device state. Indicates that the real-time data of the device is normal and the action =1 means that the real-time data of the device is judged to be abnormal; Define a reward function. When the agent correctly judges the device status, it will be given a positive reward. When it makes an incorrect judgment, it will be given a negative reward. Set the device actual status to , Indicates normal, Indicates abnormality; Then the agent is trained through the Q learning algorithm to obtain the reinforcement learning agent.

9. The method for analyzing equipment fault detection data for a digital factory according to claim 1, characterized in that: The process of training an agent through the Q learning algorithm to obtain a reinforcement learning agent is as follows: Determine the agent's state, action space, and reward function, initialize the Q network parameters and experience replay buffer, and set the total training time step and discount factor; At each time step, observe the state of the agent, through The greedy strategy selects an action, obtains the reward and new state of the environment feedback after execution, and stores them in the experience replay buffer; Randomly extract samples from the experience replay buffer and obtain the Q value of the target, calculate it at the same time, and then update the Q network parameters through the back propagation algorithm; The loss gradually decreases during training, and the Q network training steps are repeated to reach the total training time step until the Q learning network converges and the training is completed.

10. An equipment fault detection data analysis system for a digital factory, using any method described in claims 1 to 9, characterized in that: include: Synchronous acquisition module: uses high-precision clock synchronization technology to synchronously acquire comprehensive maintenance data of equipment; State association acquisition module: The location data of the device is used as the graph node, and the functional association and data interaction data between the devices are used as the edges of the graph to build a graph neural network. The device association model is built based on the graph neural network, and the state data of the device is used as the input of the device association model to output the state association factor. Comprehensive fault detection module: The comprehensive maintenance data and state correlation factors are used as the state input of the reinforcement learning agent, and the Q learning algorithm is used to train the agent to obtain the reinforcement learning agent. The real-time state data is input into the reinforcement learning agent, the state score is output, and the abnormal state threshold is set. , when the state score is lower than the abnormal state threshold , the corresponding real-time data is marked as abnormal.

Images