Intelligent management system for after-sales maintenance work order

Abstract

The invention relates to the technical field of maintenance work order management, in particular to an intelligent management system for after-sales maintenance work orders, which realizes equipment clustering based on a K-means algorithm, considers different user habits and use environment differences, and improves the fault diagnosis precision. A fault propagation network is constructed in combination with association rule mining and a graph convolutional network, and a fault influence path between hardware is described, so that accurate identification of target fault hardware is realized, alternative hardware is intelligently determined in combination with recent maintenance records, and misjudgment and repeated maintenance are avoided. And the intelligent order dispatching module is used for preferentially dispatching the nearby maintenance personnel with the required spare parts by matching the position information of the maintenance personnel with the real-time spare part inventory, so that the maintenance response efficiency is improved. And the model error correction module continuously monitors the operation state of the equipment after maintenance is completed, automatically adjusts node features and edge weight parameters if a prediction result does not accord with actual maintenance, updates a graph convolution model through an incremental learning mode, and improves the self-adaption and evolution capability of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of maintenance work order management, and particularly to an intelligent management system for after-sales maintenance work orders. Background Art

[0002] The fault detection and after-sales maintenance of household electrical appliances have become important issues affecting user experience and after-sales service efficiency. Traditional after-sales maintenance services often rely on users to actively report repairs and manual troubleshooting by maintenance personnel, which not only has low efficiency, but also has problems such as insufficient accuracy of fault diagnosis, high repeat repair rate, and slow repair response speed. To improve the service quality and response speed of after-sales maintenance, existing technical solutions have attempted to use Internet of Things technology means to achieve preliminary intelligent dispatching and equipment status monitoring. However, existing intelligent repair solutions generally only perform fault judgment based on simple device parameter threshold warnings or general artificial intelligence classification models, ignoring the differences between device groups, fault propagation paths, and the correlation between specific hardware under different usage habits and scenarios. Therefore, there are problems such as inaccurate fault identification, frequent repeat repairs, and unreasonable repair dispatching in actual applications.

[0003] In addition, existing technologies do not fully consider the recent repair history of maintenance hardware, and there are often situations where the same hardware is repaired multiple times but the actual fault occurs in other associated hardware, resulting in increased maintenance costs and reduced user satisfaction. At the same time, the current technical solutions lack an effective model feedback and error correction mechanism. Once misdiagnosis occurs, it is difficult to automatically update the model, and the probability of continuous misjudgment is relatively high, which limits the self-adaptability and sustainable optimization ability of the system. Summary of the Invention

[0004] To solve the above problems, the present invention provides an intelligent management system for after-sales maintenance work orders.

[0005] To achieve the above object, the technical solution adopted by the present invention is: An intelligent management system for after-sales maintenance work orders, comprising: a device clustering module, a fault network model construction module, a faulty hardware location module, an intelligent dispatching module, and a model error correction module; The device clustering module is used to collect the operation parameters, environmental parameters, and error logs of a number of devices through the cloud, and perform clustering analysis on the devices based on the K-means clustering algorithm to obtain a number of device clusters; The fault network model construction module is used to establish an initial fault propagation network based on the operation parameters, environmental parameters, and error logs of the devices in the device cluster through an association rule mining algorithm, and optimize and train the node features and edge weights of the initial fault propagation network through a graph convolutional network to obtain a fault network model; The fault hardware location module is used to extract fault features based on the current operating parameters and error logs of the faulty device when a faulty device is found in real time, and determine the target fault hardware through the fault network model; The intelligent work order dispatching module is used to determine alternative faulty hardware associated with the target faulty hardware through the fault network model according to whether the target faulty hardware has been repaired or replaced recently, dispatch work orders with reference to the real-time spare part inventory of maintenance personnel within the preset distance, and update the spare part inventory information after the repair is completed; The model error correction module is used to detect the status of the monitored device after repair and correct the nodes and edge weights of the fault network model.

[0006] Further, the steps of collecting the operating parameters, environmental parameters and error logs of several devices include the following: The built-in monitoring device of the device collects parameters such as voltage, current, temperature, operating time, operating power and start-stop frequency; The environmental temperature parameter and humidity parameter are collected in real time through the environmental sensors deployed by the device; The error logs generated by the device are recorded in real time through the listening code, including error codes, occurrence times and fault description information.

[0007] Further, the steps of performing clustering analysis on the devices based on the K-means clustering algorithm include the following: Convert the operating parameters, environmental parameters and error logs of the devices into feature vectors to form a dataset of features to be clustered; Initialize the clustering centers, calculate the distance between each device sample in the dataset of features to be clustered and each clustering center, and assign it to the corresponding device cluster according to the minimum distance; Update the clustering centers based on the current device clusters, recalculate the distances between all device samples and the clustering centers, and update the assignment results; Repeat the operations of updating the clustering centers and reassigning the device clusters until the clustering centers converge and no longer change, and obtain the device cluster division results.

[0008] Further, the steps of establishing an initial fault propagation network based on the operating parameters, environmental parameters and error logs of the devices in the device clusters through the association rule mining algorithm include the following: Standardize the operating parameters, environmental parameters and error logs of each device in the device clusters to construct a structured transaction dataset; Mine frequent item sets from the transaction dataset through the Apriori algorithm or extract the high-frequency association patterns between different hardware components when a fault occurs; Construct an initial fault propagation network with hardware components as nodes and fault co-occurrence relationships as edges based on frequent itemsets, and use the support and confidence of the itemsets as the initial weights of the edges, and output the initial fault propagation network.

[0009] Further, the optimization training of the node features and edge weights of the initial fault propagation network by the graph convolutional network includes the following steps: For each hardware node in the initial fault propagation network, based on the abnormal degree of the overall machine operation parameters, the abnormal degree of the environmental parameters, and the frequency of specific error logs of the corresponding device when any hardware fault occurs in the device cluster, statistical calculations are performed to obtain a historical abnormal feature vector characterizing the hardware node, and an input feature matrix of the graph convolutional network is constructed; Construct an adjacency matrix based on the fault co-occurrence relationship between hardware nodes in the initial fault propagation network; Input the input feature matrix and the adjacency matrix into the graph convolutional network model, perform graph convolutional operations to aggregate and propagate node features, and output the target faulty hardware; Construct a labeled data set based on the historically known faulty hardware, define a loss function using supervised learning, and train and optimize the parameters of the graph convolutional network through the backpropagation algorithm; Generate a fault network model after training and optimization.

[0010] Further, the convolution operation formula of the graph convolutional network is as follows: ; Among them, is the node feature matrix of the th layer of the graph convolutional network, is the node feature matrix of the th layer of the graph convolutional network. When is 0, is the input feature matrix; is the matrix after adding self-connections to the k-th order adjacency matrix; is the corresponding degree matrix; is the trainable weight matrix for the k-th order propagation of the th layer; is the trainable weight matrix for the residual connection path of the th layer; is the k-th order adjacency propagation weight coefficient; is the total order; is the non-linear activation function.

[0011] Further, the extraction of fault features based on the current operating parameters and error logs of the faulty device, and the determination of the target faulty hardware by the fault network model include the following steps: For a device with a real-time fault, collect its current operating parameters and environmental parameters, and extract the current error types and frequencies from the error log; According to the historical means of the normal operating parameters and environmental parameters of the device cluster to which the device belongs, calculate the abnormal deviation degree of the current device's operating parameters and environmental parameters, and form the feature vector of the current faulty device; Convert the feature vector of the current faulty device into the input feature matrix of the current faulty device, input it into the trained and optimized fault network model, and calculate the fault probabilities of each hardware node through graph convolutional network propagation; Output the hardware node with the highest fault probability as the target faulty hardware.

[0012] Further, the steps of determining the alternative faulty hardware associated with the target faulty hardware through the fault network model according to whether the target faulty hardware has been repaired or replaced recently are as follows: After determining the target faulty hardware, query the recent repair or replacement records of the target faulty hardware, and judge whether the target faulty hardware has been repaired or replaced within the preset period; If the target faulty hardware has been repaired or replaced, obtain the associated hardware nodes that have a direct edge connection with the target faulty hardware and the edge weight is higher than the preset threshold through the fault network model; Sort according to the fault probabilities of the associated hardware nodes from high to low, and select the top two associated hardware as the alternative faulty hardware for output.

[0013] Further, the steps of dispatching work orders according to the real-time spare part inventory of maintenance personnel within the preset distance and updating the spare part inventory information after the repair are completed are as follows: According to the location of the device with a real-time fault found, obtain the location information and real-time spare part inventory information of maintenance personnel within its preset distance range; If more than one qualified maintenance personnel are screened out, sort according to the distance between the maintenance personnel and the faulty device, and dispatch a work order to the maintenance personnel closest to the faulty device; After the maintenance personnel complete the repair, update the spare part inventory information of the maintenance personnel in real time through the mobile terminal, and upload the updated spare part inventory data for synchronous update.

[0014] Further, the model error correction module is used to perform the following steps: After the repair is completed, collect the operating parameters, environmental parameters and error log of the device in real time, and monitor the operating status of the device within the preset time period; If the device does not encounter the same type of error or exception again within the preset time period, and the actually repaired hardware is inconsistent with the target faulty hardware predicted by the fault network model, then execute S1 - S2; if the device still encounters the same type of error or exception within the preset time period, or the actually repaired hardware is consistent with the target faulty hardware predicted by the fault network model, then do not perform any operation; S1. Based on the node positions of the actually repaired hardware and the target faulty hardware in the network, correct the feature vectors of the target faulty hardware nodes, including reducing the abnormal degree value in the feature vector corresponding to the target faulty hardware node and increasing the abnormal degree value in the feature vector corresponding to the actually repaired hardware node; S2. According to the current prediction deviation, directionally adjust the edge weights of the adjacency matrix between the target faulty hardware and the actually repaired hardware nodes, including reducing the edge weights related to the target faulty hardware and increasing the edge weights related to the actually repaired hardware nodes.

[0015] The beneficial effects of the present invention are as follows: By collecting the operating parameters, environmental parameters, and error logs of household appliances, accurately dividing the device groups through the K - means clustering algorithm, considering the differences in different user habits and usage scenarios. Then, based on the characteristic data of the device groups, using the association rule mining algorithm and the graph convolutional network to accurately construct the device fault propagation network, effectively identifying the association relationships between hardware and the fault propagation paths, and achieving high - precision target hardware fault location. Combining the recent repair history of the target faulty hardware, intelligently determining alternative faulty hardware through the fault network model, avoiding repeated repair of the same hardware, and optimizing the dispatch decision. The intelligent dispatch module further intelligently matches the most suitable repair personnel according to the real - time location information and the spare part inventory situation of the repair personnel, minimizing the response time to the greatest extent and improving the repair efficiency. At the same time, this solution sets up a model error - correction module, which automatically corrects the model feature vectors and edge weight parameters by monitoring the operating status of the device after repair, continuously improving the model accuracy and adaptability, and effectively reducing the model misjudgment rate. Compared with traditional after - sales repair services, this solution significantly improves the fault diagnosis accuracy and repair response speed, reduces the number of repeated repairs, lowers the repair cost, effectively improves the user experience and satisfaction, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic structural diagram of an intelligent management system for after - sales repair work orders in the present invention.

[0017] Figure 2 is a flow chart of the steps for optimizing and training the node features and edge weights of the initial fault propagation network through the graph convolutional network in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Please refer to Figure 1 - Figure 2As shown, the present invention relates to an intelligent management system for after-sales maintenance work orders, including: an equipment clustering module, a fault network model construction module, a faulty hardware location module, an intelligent work order dispatching module, and a model error correction module; The equipment clustering module is used to collect the operating parameters, environmental parameters, and error logs of a number of devices through the cloud, and perform clustering analysis on the devices based on the K-means clustering algorithm to obtain a number of device clusters; The fault network model construction module is used to establish an initial fault propagation network based on the operating parameters, environmental parameters, and error logs of the devices in the device cluster through an association rule mining algorithm, and optimize and train the node features and edge weights of the initial fault propagation network through a graph convolutional network to obtain a fault network model; The faulty hardware location module is used to extract fault features based on the current operating parameters and error logs of a faulty device when a faulty device is detected in real time, and determine the target faulty hardware through the fault network model; The intelligent work order dispatching module is used to determine alternative faulty hardware associated with the target faulty hardware through the fault network model according to whether the target faulty hardware has been repaired or replaced recently, dispatch work orders in accordance with the real-time spare part inventory of maintenance personnel within a preset distance, and update the spare part inventory information after the repair is completed; The model error correction module is used to detect the status of the monitored device after the repair and correct the nodes and edge weights of the fault network model.

[0019] In some embodiments, first, the device clustering module is responsible for collecting the operating parameters of household appliances such as air conditioners, washing machines, refrigerators, etc., such as current, voltage, temperature, environmental parameters such as indoor temperature and humidity, and error logs such as startup failures and overload protection. After the system maps the above raw data into high-dimensional feature vectors using feature engineering methods, it performs unsupervised classification of the devices through the K-means clustering algorithm. The determination of the K value is evaluated by combining the silhouette coefficient and the elbow method to ensure the compactness and separability of the clustering results in terms of category division. The finally obtained device clusters will be used to construct a more similar fault propagation network structure basis. Next, the fault network model construction module first mines the frequent itemset relationships between faulty components in the structured transaction dataset using the Apriori algorithm based on the data distribution within the device clusters, and constructs an initial fault propagation network based on this, where the nodes represent specific hardware components, the edges represent the co-occurrence relationships between their faults, and the edge weights are jointly determined by the support and confidence. Subsequently, this network structure and the historical anomaly statistical features corresponding to each node (hardware) (such as the fault frequency of a component in the device cluster, the average deviation value of the corresponding device parameters, and the occurrence frequency of associated error logs) are jointly used as the input of the graph convolutional network. The graph convolutional network (GCN) adopts a multi-layer propagation structure to achieve the aggregation and propagation of node features. During the training process, the labels of the target hardware nodes are optimized through supervised learning, and the loss function is designed as cross-entropy plus L2 regularization term to suppress overfitting, and finally outputs a fault network model with reasoning ability. When the system detects that a certain device has a fault, the faulty hardware location module will collect its operating parameters and error logs in real time and extract the current state features, generate the current input feature vector by calculating the deviation degree between the state features and the historical distribution of the device cluster, and input it into the trained fault network model. The GCN outputs the fault probability distribution of each node (i.e., each hardware component) through multi-layer adjacency matrix propagation and node weight iteration. The system finally determines the target faulty hardware as the one with the highest probability. After obtaining the target faulty hardware, the intelligent dispatch module first determines whether the hardware has been repaired or replaced within the preset time range. If so, it indicates that there may be misjudgment or it is a secondary fault. The system will call the fault network model to output its adjacent nodes with high edge weights as alternative faulty hardware, and sort them in combination with their current fault probabilities. Subsequently, the system obtains the real-time location information of all maintenance personnel within the geographical location range and the list of spare parts they carry, and jointly calculates the maintenance dispatch priority through the spatial distance and inventory matching function, and gives priority to dispatching work orders to the personnel who carry the target spare parts and are closer, improving the response efficiency and reducing resource waste. After the repair is completed, the system will automatically collect and upload the used spare parts and the current remaining inventory through the mobile terminal, and update the central inventory information database in real time.To enhance the system's self-learning ability and prediction reliability, the model error correction module conducts state backtracking monitoring on all devices with completed repairs. If the target faulty hardware predicted by the system is inconsistent with the actual repaired hardware, and the device does not exhibit recurrence anomalies within the preset time window, it is regarded as a valid error correction signal. The system will automatically lower the anomaly intensity of the original prediction node, enhance the anomaly characteristics of the actual repair node, and trigger incremental training of the graph convolutional network based on the current corrected samples to update the node weights and edge weight structure, enabling the model to gradually converge to the true fault logic. Through the collaborative operation of the above modules, this system significantly outperforms existing solutions in terms of equipment fault identification accuracy, work order dispatching efficiency, and model adaptive optimization ability. The equipment clustering module constructs a differentiated foundation, enhancing the generalization ability of the model; the fault network model depicts the fault logic path, realizing prediction and reasoning from a global perspective; the intelligent dispatching and inventory linkage reduces the scheduling cost, forming a data closed-loop feedback among the modules.

[0020] Further, the steps of collecting the operation parameters, environment parameters, and error logs of several devices include the following: The built-in monitoring device of the device collects parameters such as voltage, current, temperature, operation time, operation power, and start-stop frequency; The environmental temperature parameter and humidity parameter are collected in real time through the environmental sensors deployed on the device; The error logs generated by the device are recorded in real time through the listening code, including error codes, occurrence times, and fault description information.

[0021] Specifically, the system first constructs a multi-source data acquisition architecture, supporting high-frequency and high-precision acquisition of device operation parameters, environment parameters, and error logs. Taking an embedded intelligent air conditioner device as an example, the built-in operation monitoring module of the device includes multiple analog signal acquisition channels and a digital signal processing unit. The air conditioner device contains multiple hardware components such as a main control chip, a power module, a compressor, and a motor. Specifically, the system collects the voltage, current, and temperature parameters of the device through a high-precision sensor module. Among them, voltage and current can be used to reflect the load state and power consumption, while temperature can monitor whether there are hidden thermal protection hazards. In addition, the system also periodically reads the operation time and start-stop frequency, which are respectively used to judge the continuous working duration and frequent start-stop behavior of the device, so as to assist in judging whether there are operation anomalies.

[0022] At the environmental perception level, the system relies on the environmental sensor unit deployed around the device to collect the environmental temperature and relative humidity of the space where the device is located in real time. This part of the data is crucial for determining the cause of the fault. For example, the overheating anomaly of the compressor may be caused by poor heat dissipation due to too high indoor temperature or too low humidity. The combined analysis of the change trend of environmental parameters and the operation behavior of the device can provide an important background benchmark for subsequent clustering and anomaly identification.

[0023] In addition, at the software level, the system realizes the collection of abnormal event logs during the operation of the device through an embedded monitoring code module. The log content includes error codes, the time of fault occurrence, and the fault description information returned by the system. This monitoring module is integrated into the device operating system or application layer program, and can automatically record event information and push it to the cloud platform when a fault event is detected, constituting a key data source for sequential fault records. The error codes follow a unified specification coding standard and can be directly mapped to the component identifiers in the device structure diagram, which is used for subsequent annotation and label extraction of the fault network model.

[0024] Furthermore, the clustering analysis of the device based on the K-means clustering algorithm includes the following steps: Convert the operating parameters, environmental parameters, and error logs of the device into feature vectors to form a dataset of features to be clustered; Initialize the clustering centers, calculate the distance between each device sample in the dataset of features to be clustered and each clustering center, and assign it to the corresponding device cluster according to the minimum distance; Update the clustering centers based on the current device clusters, recalculate the distances between all device samples and the clustering centers, and update the assignment results; Repeat the operations of updating the clustering centers and reassigning the device clusters until the clustering centers converge and no longer change, and obtain the result of device cluster division.

[0025] In a specific embodiment, taking a certain brand of smart washing machine as an example, its original collected data includes operating parameters (such as current, voltage, drum motor load, heating power, program duration, etc.), environmental parameters (such as indoor temperature, humidity), and error logs (such as abnormal water inlet, dehydration failure, door lock abnormality, etc.). To achieve numerical unity and accuracy of feature representation, the system uses the Z-score standardization method to normalize continuous parameters to ensure comparability between different dimensions. At the same time, the error logs are converted into sparse vectors through one-hot encoding, and then the discriminative ability of fault information for the clustering process is enhanced through TF-IDF (Term Frequency-Inverse Document Frequency) weights. Finally, each device is represented as a set of feature vectors with a fixed dimension, forming a dataset of features input to the K-means algorithm.

[0026] In the initialization stage of the K-means clustering algorithm, the system selects the initial clustering centers through the K-means++ strategy to reduce the local optimum problem caused by random initialization. For each device sample in the feature dataset, the system uses the Euclidean Distance as the similarity metric, calculates its distances from all clustering centers, and assigns the sample to the device cluster represented by the nearest clustering center. After the initial clustering assignment is completed, the system calculates the mean of the member sample features in each device cluster and updates the coordinates of the clustering centers. Subsequently, the distances between all samples and the new clustering centers are recalculated, and the cluster membership is re-divided. This process is carried out iteratively until the change in the positions of all clustering centers in two consecutive iterations is less than the preset convergence threshold or the maximum number of iterations is reached, thereby outputting the final device cluster division result. In actual deployment, the system sets appropriate K values for different device categories. The selection of the K value comprehensively considers the results of the Silhouette Coefficient, the Calinski-Harabasz index, and the Elbow method, and offline tuning is completed before deployment. The finally output device clusters not only improve the accuracy of subsequent graph neural networks in modeling fault relationships but also serve as the structural support for upper-layer functions such as maintenance cycle prediction and fault mode migration analysis under differential strategies.

[0027] Further, establishing an initial fault propagation network based on the operating parameters, environmental parameters, and error logs of the devices in the device cluster through an association rule mining algorithm includes the following steps: Standardize the operating parameters, environmental parameters, and error logs of each device in the device cluster to construct a structured transaction dataset; Mine frequent item sets from the transaction dataset through the Apriori algorithm or extract the high-frequency association patterns between different hardware components when faults occur; Based on the frequent item sets, construct an initial fault propagation network with hardware components as nodes and fault co-occurrence relationships as edges, and use the support and confidence of the item sets as the initial weights of the edges to output the initial fault propagation network.

[0028] It should be noted that first, the operating parameters, environmental parameters, and error logs of each device are uniformly encoded and standardized. Among them, the operating parameters and environmental parameters adopt the interval discretization strategy, and the continuous numerical variables are divided into discrete labels such as "normal", "high", or "low" according to the statistical distribution within the device cluster to enhance the identifiability of transaction items. For example, if the average current in a device cluster is 2.5 A and the standard deviation is 0.3 A, then when the recorded value of a certain device is 3.1 A, it will be marked as "high current" in this dimension. The error logs are transformed into independent fault event labels based on the error types (such as E01 - temperature control failure, E02 - fan abnormality). The above various labels together constitute the transaction item set of the device, thus obtaining a complete structured transaction data set in the device cluster. After the transaction data set is constructed, the system uses the Apriori algorithm to mine frequent item sets from the transaction set. The Apriori algorithm improves the calculation efficiency while ensuring the coverage of frequent item sets through an iterative pruning method. In each iteration, the system evaluates the support and confidence metrics of the item sets to screen out the combinations of fault events that frequently occur simultaneously. For example, in the transactions of multiple devices, if "high compressor current" and "main control board fault alarm" frequently occur simultaneously and meet the set thresholds of support higher than 10% and confidence higher than 60%, then this combination will be recognized as an effective fault association rule. Based on the mined frequent item sets, the system constructs an initial fault propagation network graph. Each node in the network represents a specific hardware component or its corresponding typical fault event, and the edges between the nodes represent the co-occurrence or potential causal relationship of the fault events. The initial assignment of the edge weights is based on the support and confidence of the corresponding frequent item sets. The higher the support, the more common the fault mode is in the device cluster, and the higher the confidence, the more likely it is that one fault will cause another fault, thus enhancing the authority and reliability of the propagation path. Taking a certain electric water heater device cluster as an example, in the transaction mining, it is found that the co-occurrence frequency of "low temperature sensor reading" and "heating tube abnormality alarm" is relatively high, with a support of 12% and a confidence of 78%. The system establishes a directed edge between the "temperature sensor node" and the "heating tube node", and the initial edge weight is set to the weighted function value of this group of data (such as support × confidence) for the subsequent training and optimization of the graph convolutional network model.

[0029] Furthermore, the optimization training of the node features and edge weights of the initial fault propagation network by the graph convolutional network includes the following steps: For each hardware node in the initial fault propagation network, based on the abnormal degree of the overall operating parameters of the corresponding device, the abnormal degree of the environmental parameters, and the frequency of specific error logs when any hardware fault occurs in the device cluster, statistical calculations are performed to obtain a historical abnormal feature vector representing the hardware node, and an input feature matrix of the graph convolutional network is constructed. Construct an adjacency matrix based on the fault co-occurrence relationship between hardware nodes in the initial fault propagation network; Input the input feature matrix and the adjacency matrix into the graph convolutional network model, perform graph convolutional operations to aggregate and propagate node features, and output the target faulty hardware; Construct an annotation dataset based on historically known faulty hardware, define a loss function using supervised learning, and train and optimize the graph convolutional network parameters through the backpropagation algorithm; Generate a fault network model after training and optimization.

[0030] In some embodiments, the system optimizes and trains the node features and edge weights in the initial fault propagation network through a graph convolutional network to improve the accuracy of fault identification and the generalization ability of the model. Specifically, first, a feature representation is constructed for each hardware node in the initial fault propagation network. This feature vector does not come from data collection of the hardware itself, but by statistically analyzing the abnormal degree of the overall machine operation parameters, the deviation degree of environmental parameters, and the occurrence frequency of the error logs corresponding to the hardware fault when the hardware fault occurs in the device cluster where it is located. Taking an air conditioner compressor as an example, the system screens out the sample devices that have been maintained due to compressor faults in the device cluster, and extracts the deviation of the operating current, voltage, and power, the change trend of the indoor and outdoor temperature difference, and the concentration degree of relevant error codes of these devices before the fault occurs. Through numerical and standardized processing of these abnormal features, a feature vector representing the historical abnormal performance of the hardware is formed. Then, the system constructs an adjacency matrix according to the fault co-occurrence relationship between nodes in the initial fault propagation network. Each edge represents that two hardware components have frequently co-occurred faults in historical devices, and its weight is determined by the support and confidence obtained from frequent item set mining. This adjacency matrix not only describes the association structure between hardware, but also is the core basis for subsequent feature aggregation and propagation of the graph neural network. During the training process, the system inputs the above-constructed node feature matrix and adjacency matrix into the graph convolutional network. The network realizes feature enhancement by aggregating the information of node neighbors layer by layer, so that the final representation of each node not only contains its own historical abnormal information, but also integrates the fault influence of its associated nodes. Through the supervised learning method, combined with the fault label data set constructed based on historical maintenance information, the network is guided to train the ability to predict fault nodes. The loss function uses cross-entropy to minimize the error between the fault prediction result and the actual label, and the backpropagation and gradient descent algorithms are used in the training process for parameter iterative update. This method is significantly different from the prior art in many aspects. On the one hand, the system does not rely on the traditional method of judging whether there is a fault based on the overall threshold of the device, but based on the group behavior patterns between devices and components, establishes a graph network model with structure recognition ability; on the other hand, by introducing structural dependence through the graph convolutional neural network, the system still has a certain reasoning ability when the data is incomplete or the signal of a single hardware is not significant, effectively improving the diagnostic accuracy in complex scenarios. Through the deep fusion of graph structure information and node features, this embodiment realizes the efficient integration and cross-modeling of multi-source information, further enhances the context sensitivity of fault identification, and forms a high-dimensional reasoning model different from the traditional flat processing method.

[0031] Further, the convolution operation formula of the graph convolutional network is as follows: ; Where is the The node feature matrix of the layer graph convolutional network, is the node feature matrix of the -th layer graph convolutional network. When is 0, is the input feature matrix; is the matrix after adding self-connections to the k-th order adjacency matrix; is the corresponding degree matrix; is the trainable weight matrix for the k-th order propagation of the -th layer; is the trainable weight matrix for the residual connection path of the -th layer; is the weight coefficient for the k-th order adjacency propagation; is the total order; is the non-linear activation function.

[0032] It should be noted that is the node feature matrix of the -th layer of the graph convolutional network. Each row corresponds to a hardware component node, and each column is a feature value of the node. In this solution, the initial is a vector matrix constructed based on statistical features such as the abnormal degree of the overall machine operation parameters, the abnormal degree of environmental parameters, and the error log frequency when the component represented by the node appears in the overall device. is the node feature matrix of the -th layer of the graph convolutional network, which is the output of the current layer and the input of the next layer. This matrix reflects the aggregation result of node features after considering neighbor information. is the k-th order adjacency matrix, which represents the connection relationship between each node in the graph structure and its k-th order neighbors in matrix form, and adds self-connections (i.e., adding 1 to the diagonal) so that each node retains its own information during propagation. In this solution, the 0-th order represents the current node, the 1-st order represents the directly connected hardware components, and the 2-nd order represents the indirect fault path propagated through an intermediate hardware. is the k-th order adjacency matrix 's degree matrix, which is a diagonal matrix corresponding to the number of connections (i.e., the number of neighbors) of each node. It is used to normalize , preventing nodes with a large number of connections from having too large weights during propagation. is the propagation weight coefficient of the k-th order adjacency path, which is trainable and used to control the contribution degree of different order propagations to the final feature. In this solution, this parameter reflects the different importance of fault propagation on direct and indirect paths. For example, when the first-order adjacency propagation is more representative than the second-order propagation, the system can automatically learn to increase the weight. Control the transformation of neighbor information in the feature dimension. During actual training, the system will automatically learn the optimal weights according to the fault propagation patterns between different hardware in historical data. It directly applies a linear transformation to the input features and then superimposes the result on the convolution result. The residual connection helps to retain the original features, alleviate the vanishing gradient problem in deep neural networks, and improve the training stability. is the ReLU non-linear activation function, which is used to introduce non-linear feature transformation and enable the model to have stronger fitting ability.

[0033] Furthermore, the steps of extracting fault features based on the current operating parameters and error logs of the faulty device and determining the target faulty hardware through the fault network model are as follows: For the device with a real-time fault, collect its current operating parameters and environmental parameters, and extract the current error types and frequencies from the error logs; Calculate the abnormal deviation degree of the current device's operating parameters and environmental parameters according to the historical mean values of the normal operating parameters and environmental parameters of the device cluster to which the device belongs, and form the feature vector of the current faulty device; Convert the feature vector of the current faulty device into the input feature matrix of the current faulty device, input it into the trained and optimized fault network model, and calculate the fault probabilities of each hardware node through the graph convolution network propagation; Output the hardware node with the highest fault probability as the target faulty hardware.

[0034] Specifically, first, the system automatically retrieves the operation parameters and environmental parameter data of the device at the time of the fault occurrence. The operation parameters include, but are not limited to, real-time voltage, current, temperature, power, operation time, and start-stop frequency; the environmental parameters mainly include the temperature and humidity in the device's working environment. At the same time, the system synchronously extracts the error log data recorded at the moment of the device fault occurrence, including the type of error code, the frequency of repeated occurrences, and its time series distribution characteristics. Next, the system compares the currently collected operation parameters and environmental parameters of the faulty device with the mean and variance of the operation parameters of its device cluster under the historical normal state. By calculating statistical deviation indicators such as Z-score, the abnormality degree of the current state compared with the average state of the same type of device group is quantified, thereby constructing a multi-dimensional feature vector representing the current fault state of the device. This feature vector not only contains the numerical offset of the operation indicators, but also integrates the reflection of the impact of environmental fluctuations on the device performance, as well as the time frequency pattern of the log events, ensuring the formation of a timely and discriminative input data structure. The above feature vector will be encoded into the input feature matrix of the current faulty device and input into the trained graph convolutional network model. The graph convolutional network is a fault propagation graph constructed based on historical multi-device multi-fault collaborative data. Each node in the graph represents a hardware component, and the edge weight represents its co-occurrence probability with other hardware in the fault event. The graph convolutional operation combines the feature vector of the current faulty device with the structural connection relationship of each node in the graph, performs layer-by-layer information propagation and aggregation, and finally generates the fault probability prediction for each hardware node at the output layer. The system sorts the fault probabilities of all hardware nodes and selects the node with the highest probability as the target faulty hardware. This fault location method can comprehensively consider the manifestation form of the current device state abnormality and the commonality of the fault modes among historical multi-devices, and can accurately judge the internal fault source without separately monitoring each hardware component with sensors. The method adopted in this embodiment has significant advantages compared with the existing scheme that relies on static thresholds and single classification output based on the full-device model. By comparing and normalizing the current device state with the evolution characteristics of a large-scale device cluster and then embedding it into the graph structure model for correlation reasoning, not only the accuracy of fault location is improved, but also the adaptability of the system to the complex and multi-source heterogeneous device operation environment is enhanced, reflecting the innovation and practical value of the present invention in terms of algorithm structure and application effect.

[0035] Further, the steps of determining the alternative faulty hardware associated with the target faulty hardware through the fault network model according to whether the target faulty hardware has been repaired or replaced recently include the following: After determining the target faulty hardware, query the recent repair or replacement record of the target faulty hardware to judge whether the target faulty hardware has been repaired or replaced within the preset period; If the target faulty hardware has been repaired or replaced, then the fault network model is used to obtain associated hardware nodes that are directly connected to the target faulty hardware and whose edge weights are higher than a preset threshold; The associated hardware nodes are sorted according to their failure probabilities, and the top two associated hardware are selected as candidate fault hardware outputs.

[0036] Specifically, after the system preliminarily determines the target faulty hardware node based on the output of the fault network model, it will call the equipment historical maintenance database to retrieve the maintenance and replacement records of the hardware within the preset time window (such as the past 30 days). If it is found that the target faulty hardware has been replaced or repaired recently, the system will further start the association reasoning process of the candidate hardware to deal with the risk of "invalid replacement" or "misjudgment of the main cause". In the process of identifying associated hardware, the system first obtains other hardware nodes that have direct edge connections with the target faulty hardware node in the fault network model and reads the weight value of the edge. The edge weight has been optimized during the graph convolution training process and can effectively characterize the strength of the linkage between the failures of the two hardware components in historical samples. The system compares the edge weight with the preset threshold (for example, 0.6) and only retains the hardware nodes with edge weights greater than the threshold as the high-association hardware candidate set of the current target faulty hardware. Subsequently, the system re-evaluates the failure probability score of the candidate hardware node based on the fault network model under the input characteristics of the current faulty device. This evaluation utilizes the aforementioned graph convolution propagation mechanism, that is, the candidate node will calculate the probability of possible failure in the current scenario based on the similarity between it and the current feature matrix of the faulty device, the adjacency weight on the propagation path and its historical anomaly vector. Finally, the system outputs the top two hardware nodes in the candidate set with the highest probability of failure as candidate faulty hardware. This method solves the problem of the inability to identify replacement failures or fault transfer situations in traditional solutions by introducing the strategy of "proximate cause negation + graph association confirmation", and is particularly suitable for scenarios with high coupling between modules in complex home appliance systems. Compared with the existing dispatching logic that only relies on single-point prediction of fault probability or does not consider historical maintenance behavior, this embodiment fully reflects the advantages of combining graph structure modeling with state dynamic analysis, and realizes the closed loop of "fault prediction + misjudgment avoidance" from the perspective of algorithm collaboration and actual application effect, significantly improving the system's intelligent decision-making ability and user satisfaction.

[0037] Furthermore, the dispatching of work orders according to the real-time spare parts inventory of maintenance personnel within a preset distance and updating spare parts inventory information after the maintenance is completed includes the following steps: According to the real-time location of the faulty equipment, the location information of the maintenance personnel within the preset distance and the real-time spare parts inventory information are obtained; If more than one eligible maintenance personnel are selected, they will be sorted according to their distances from the faulty device, and a work order will be dispatched to the maintenance personnel closest to the device. After the maintenance is completed, the spare part inventory information of the maintenance personnel will be updated in real time through the mobile terminal, and the updated spare part inventory data will be uploaded for synchronous update.

[0038] In some embodiments, after the faulty hardware or alternative faulty hardware is determined, the system first determines the spatial location information of the faulty device based on the location coordinate information uploaded by the device terminal, in combination with the GIS positioning service or the deployed geocoding module. Subsequently, the system retrieves the real-time location information of all current maintenance personnel and the spare part inventory list uploaded by their mobile terminals or the system. The inventory data is dynamically synchronized through the mobile terminal with the intelligent spare part boxes carried by the personnel, including the quantity, model, and batch information of various hardware modules, and can be uploaded to the cloud in real time for unified management. The system filters out the maintenance personnel within the preset service radius (such as 5 kilometers, 8 kilometers, etc.), and determines whether they carry the target faulty hardware and its alternatives. If multiple maintenance personnel meet the conditions, they will be further sorted according to their current spatial distance from the faulty device, and the one closest to the device will be selected first to dispatch the work order, minimizing the response time and avoiding resource waste. After the maintenance task is completed, the maintenance personnel enter the types and quantities of the hardware actually replaced during this maintenance through the mobile terminal, and the system will deduct the local inventory records of this maintenance personnel in real time. The updated inventory data is simultaneously synchronized to the platform background for inventory verification and replenishment warning management for subsequent tasks, ensuring the consistency and timeliness of the inventory data. Through the above implementation path, this module realizes the dual-factor dispatching optimization strategy based on location constraints and inventory matching, which is different from the extensive mechanism of only using regional division or manual assignment for task scheduling in the prior art, and has the advantages of strong real-time performance, high decision-making accuracy, and high resource utilization rate, effectively improving the overall operation efficiency of after-sales maintenance and customer satisfaction.

[0039] Further, the model error correction module is used to perform the following steps: After the maintenance is completed, the operating parameters, environmental parameters, and error logs of the device are collected in real time, and the operating status of the device within a preset time period is monitored. If the device does not encounter the same type of error or abnormality again within the preset time period, and the actually repaired hardware is inconsistent with the target faulty hardware predicted by the faulty network model, then perform S1-S2; if the device still encounters the same type of error or abnormality again within the preset time period, or the actually repaired hardware is consistent with the target faulty hardware predicted by the faulty network model, then do not perform any operation. S1. Based on the node positions of the hardware for actual repair and the target faulty hardware in the network, perform error correction on the eigenvectors of the target faulty hardware node, including reducing the abnormal degree value in the eigenvector corresponding to the target faulty hardware node and increasing the abnormal degree value in the eigenvector corresponding to the actual repair hardware node; S2. Directionally adjust the edge weights of the adjacency matrix between the target faulty hardware and the actual repair hardware node according to the current prediction deviation, including reducing the edge weights related to the target faulty hardware and increasing the edge weights related to the actual repair hardware node.

[0040] First, at the level of node eigenvectors, the system weakens the abnormal representation of the target faulty hardware node. This operation is completed by reducing the relevant components representing the degree of abnormality in the node features. Specifically, the system will find the abnormal dimension features formed by the node during the training process, such as the parameter dimensions with a high failure frequency or highly correlated with abnormal error logs in historical statistics. The system will reduce the feature values of these dimensions by a certain proportion to weaken the abnormal information propagated in subsequent graph convolution operations, thereby reducing the probability of this node being misjudged as the fault source again. At the same time, the system will also enhance the corresponding abnormal features of the actually repaired hardware node to make up for the abnormal paths that the original model failed to focus on correctly. Second, at the level of the graph structure, that is, in terms of the edge weight processing of the adjacency matrix, the system will dynamically adjust the edge weights between the target faulty hardware node and other nodes in the original model. If there is a connection relationship between the original predicted node and the actual repair node, the system will reduce the edge weight coefficient between the two to weaken its propagation influence; at the same time, increase the edge weight between the actual repair node and its associated hardware nodes to re-strengthen the structure that is more likely to form a fault propagation path. This directional edge weight fine-tuning not only corrects the local propagation relationship but also enables the graph structure to evolve continuously with the actual repair feedback after training, gradually approaching the real fault logic.

[0041] In some specific embodiments, first, the system jointly evaluates the current prediction deviation degree based on "whether the actual maintenance node appears in the initial prediction ranking" and "the activation frequency corresponding to this node in the historical samples". The system sets a three - level standard: Significant deviation level: If the actual maintenance node does not appear in the top 10 of the prediction, and its activation frequency in the historical training data is lower than 10%, it is considered that this prediction seriously deviates from the true fault mode. At this time, the abnormal degree weight of each dimension in the feature vector of the target faulty hardware node needs to be overall reduced by 30%, and at the same time, the weight of the core abnormal dimension corresponding to the actual maintenance node is increased by 40%; Moderate deviation level: If the actual maintenance node is in the 5th to 10th place in the prediction, or has been partially misdiagnosed and activated in the historical samples, the system regards it as a moderate deviation. At this time, a milder adjustment strategy is executed. The weight of the target node dimension is reduced within the range of 15% - 20%, and the actual maintenance node is enhanced by 20% - 25%; Slight deviation level: If the actual maintenance node is second only to the original target node or has an associated edge relationship with it, it is regarded as a slight deviation, and only a small - scale fine - tuning within 10% of the error - related dimension is performed. The innovation of this multi - level adjustment mechanism lies in that it not only refers to the direct difference between the current prediction result and the actual maintenance, but also combines the credibility of the node historical features in the training samples, ensuring that the adjustment strategy has both punishment and compensation, and has the stability of long - term iterative optimization. In terms of edge weight adjustment, the system further introduces a "correction edge gradient weight function", using the prediction deviation degree between the two nodes in the edge connection as the driving factor. If there is a connection edge between the target node and the actual maintenance node, the weight of this edge will be increased by 15% - 25%, and the specific increase value is proportional to the contribution rate of the edge in the historical propagation path; conversely, if there is a high - weight edge between the target node and other irrelevant nodes, the system will reverse - reduce the edge weight according to its historical propagation success rate, and the reduction range is set at 10% - 30%.

[0042] In summary, the error correction module of this embodiment realizes the dynamic adaptive update of the fault propagation model through the method of "feature correction + structure adjustment", breaking through the limitations of static model prediction and ineffective utilization of misdiagnosis feedback in the prior art, enabling the system to have the ability of continuous learning and evolution, and thus realizing more efficient and accurate intelligent decision - making support in the after - sales maintenance scenario.

[0043] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, C language, VHDL language, Verilog language, object-oriented programming language Java, and interpreted scripting language JavaScript, etc.

[0044] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks.

[0045] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions specified in Figure 1 one or more of the flows Figure 1 or blocks.

[0046] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks.

[0047] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent management system for after-sales maintenance work orders, characterized in that, Including: An equipment clustering module, a fault network model construction module, a fault hardware location module, an intelligent work order dispatching module, and a model error correction module; The equipment clustering module is used to collect the operation parameters, environmental parameters, and error logs of a number of devices through the cloud, perform clustering analysis on the devices based on the K-means clustering algorithm, and obtain a number of device clusters; The fault network model construction module is used to establish an initial fault propagation network based on the operation parameters, environmental parameters, and error logs of the devices in the device cluster through the association rule mining algorithm, and optimize and train the node features and edge weights of the initial fault propagation network through the graph convolutional network to obtain a fault network model; The fault hardware location module is used to extract fault features based on the current operation parameters and error logs of the faulty device when a faulty device is discovered in real time, and determine the target faulty hardware through the fault network model; The intelligent work order dispatching module is used to determine alternative faulty hardware associated with the target faulty hardware through the fault network model according to whether the target faulty hardware has been repaired or replaced recently, dispatch work orders in accordance with the real-time spare part inventory of maintenance personnel within a preset distance, and update the spare part inventory information after the repair is completed; The model error correction module is used to detect the status of the monitored device after the repair and correct the nodes and edge weights of the fault network model.

2. The intelligent management system for an after-sales repair work order according to claim 1, wherein The step of collecting the operation parameters, environmental parameters, and error logs of a number of devices includes the following steps: Collect the voltage, current, temperature, operation time, operation power, and start-stop frequency parameters collected by the built-in monitoring device of the device; Collect the environmental temperature parameter and humidity parameter in real time through the environmental sensors deployed on the device; Record the error logs generated by the device in real time through the listening code, including error codes, occurrence times, and fault description information.

3. The intelligent management system for after-sales repair work orders according to claim 1, characterized in that, The step of performing clustering analysis on the devices based on the K-means clustering algorithm includes the following steps: Convert the operation parameters, environmental parameters, and error logs of the devices into feature vectors to form a dataset of features to be clustered; Initialize the clustering centers, calculate the distance between each device sample in the dataset of features to be clustered and each clustering center, and assign it to the corresponding device cluster according to the minimum distance; Update the clustering centers based on the current device clusters, recalculate the distances between all device samples and the clustering centers, and update the assignment results; Repeat the operations of updating the clustering centers and reassigning the device clusters until the clustering centers converge and no longer change, and obtain the device cluster division result.

4. The intelligent management system for after-sales maintenance work orders according to claim 1, characterized in that, The step of establishing an initial fault propagation network based on the operation parameters, environmental parameters, and error logs of the devices in the device cluster through the association rule mining algorithm includes the following steps: Perform standardization processing on the operation parameters, environmental parameters, and error logs of each device in the device cluster to construct a structured transaction dataset; Mine frequent item sets from the transaction dataset through the Apriori algorithm or other algorithms, and extract the high-frequency association patterns between different hardware components when a fault occurs; Construct an initial fault propagation network with hardware components as nodes and fault co-occurrence relationships as edges based on the frequent item sets, and use the support and confidence of the item sets as the initial weights of the edges, and output the initial fault propagation network.

5. The intelligent management system for after-sales repair work orders according to claim 4, characterized in that, The optimization training of the node features and edge weights of the initial fault propagation network by the graph convolutional network includes the following steps: For each hardware node in the initial fault propagation network, based on the abnormal degree of the overall machine operation parameters, the abnormal degree of the environmental parameters, and the frequency of specific error logs of the corresponding device when any hardware fault occurs in the device cluster where it is located, statistical calculations are performed to obtain a historical abnormal feature vector characterizing the hardware node, and an input feature matrix of the graph convolutional network is constructed; Based on the fault co-occurrence relationship between the hardware nodes in the initial fault propagation network, an adjacency matrix is constructed; The input feature matrix and the adjacency matrix are input into the graph convolutional network model, and graph convolutional operations are performed to aggregate and propagate the node features, and the target faulty hardware is output; A labeled data set based on the historically known faulty hardware is constructed, a loss function is defined using supervised learning, and the parameters of the graph convolutional network are trained and optimized through the backpropagation algorithm; A trained and optimized fault network model is generated.

6. The intelligent management system for after-sales repair work orders according to claim 5, wherein The convolution operation formula of the graph convolutional network is as follows: ； Among them, is the node feature matrix of the -th layer graph convolutional network, is the node feature matrix of the -th layer graph convolutional network. When is 0, is the input feature matrix; is the matrix after adding self-connections to the k-th order adjacency matrix; is the corresponding degree matrix; is the trainable weight matrix for the k-th order propagation of the -th layer; is the trainable weight matrix for the residual connection path of the -th layer; is the k-th order adjacency propagation weight coefficient; is the total order; is the non-linear activation function.

7. An intelligent management system for after-sales maintenance work orders according to claim 5, characterized in that, The extraction of fault features based on the current operation parameters and error logs of the faulty device and the determination of the target faulty hardware through the fault network model include the following steps: For the device with a real-time fault, its current operation parameters and environmental parameters are collected, and the current error types and frequencies are extracted from the error logs; According to the historical means of the normal operation parameters and environmental parameters of the device cluster to which the device belongs, the abnormal deviation degrees of the current device operation parameters and environmental parameters are calculated to form a feature vector of the current faulty device; The feature vector of the current faulty device is converted into an input feature matrix of the current faulty device, input into the trained and optimized fault network model, and the fault probabilities of each hardware node are calculated through the propagation of the graph convolutional network; The hardware node with the highest fault probability is output as the target faulty hardware.

8. The intelligent management system for after-sales maintenance work orders according to claim 1, characterized in that, The determination of the alternative faulty hardware associated with the target faulty hardware through the fault network model according to whether the target faulty hardware has been repaired or replaced recently includes the following steps: After determining the target faulty hardware, query the recent repair or replacement records of the target faulty hardware to judge whether the target faulty hardware has been repaired or replaced within the preset period; If the target faulty hardware has been repaired or replaced, obtain the associated hardware nodes that have a direct edge connection with the target faulty hardware and the edge weight value is higher than the preset threshold through the fault network model; Sort according to the fault probability levels of the associated hardware nodes, and select the top two associated hardware as the alternative faulty hardware to output.

9. The intelligent management system for after-sales repair work orders according to claim 1, characterized in that, The dispatching of work orders according to the real-time spare part inventory of maintenance personnel within the preset distance and the updating of the spare part inventory information after the repair is completed include the following steps: According to the location of the device with a real-time fault found, obtain the location information and real-time spare part inventory information of the maintenance personnel within its preset distance range; If more than one qualified maintenance personnel are screened out, sort according to the distance between the maintenance personnel and the faulty device, and dispatch a work order to the maintenance personnel closest to the faulty device; After the maintenance personnel complete the repair, the spare part inventory information of the maintenance personnel is updated in real time through the mobile terminal, and the updated spare part inventory data is uploaded for synchronous update.

10. The intelligent management system for after-sales maintenance work orders according to claim 1, characterized in that, The model error correction module is used to perform the following steps: After the repair is completed, the operation parameters, environmental parameters, and error logs of the device are collected in real time, and the operation status of the device within a preset time period is monitored; If the device does not encounter the same type of error or anomaly again within the preset time period, and the actually repaired hardware is inconsistent with the target faulty hardware predicted by the fault network model, then execute S1-S2; if the device still encounters the same type of error or anomaly again within the preset time period, or the actually repaired hardware is consistent with the target faulty hardware predicted by the fault network model, then do not perform any operation; S1. Based on the node positions of the actually repaired hardware and the target faulty hardware in the network, the feature vector of the target faulty hardware node is error-corrected, including reducing the abnormal degree value in the feature vector corresponding to the target faulty hardware node and increasing the abnormal degree value in the feature vector corresponding to the actually repaired hardware node; S2. According to the current prediction deviation, the edge weights of the adjacency matrix between the target faulty hardware and the actually repaired hardware nodes are adjusted directionally, including reducing the edge weight of the target faulty hardware and increasing the edge weight of the actually repaired hardware node.

Images