Suspension system equipment fault diagnosis method and system

Abstract

The invention relates to a suspension system equipment fault diagnosis method and system. The method comprises the following steps: constructing a digital twinborn model of the suspension system based on a three-dimensional modeling tool; acquiring operation data of equipment in the hanging system, and binding the operation data with the digital twinborn model; extracting features representing equipment states in the operation data, analyzing the features based on an intelligent analysis model, and generating a fault diagnosis result; and generating a fault early warning signal according to the fault diagnosis result, and displaying the fault early warning signal through the digital twinborn model. According to the method, the digital twinborn model and the intelligent analysis model are used for fault diagnosis, and the problems that equipment fault detection is difficult and fault positioning efficiency is low in a traditional hanging system can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of garment intelligent manufacturing, in particular to a hanging system equipment fault diagnosis method and system. BACKGROUND

[0002] In the field of garment intelligent manufacturing, the hanging system is the core infrastructure connecting the cutting, sewing, quality inspection, warehousing and other links. Through the coordinated operation of the track, hook, sensor and control system, the automatic transmission and production process scheduling of the garment semi-finished product are realized. With the mode transformation of the garment industry, the complexity and operating load of the hanging system have significantly increased, and faults such as track deviation and bearing wear occur frequently, resulting in production line downtime, order delay and other problems. Therefore, efficient fault diagnosis has become a core requirement to ensure stable operation of the system.

[0003] The traditional garment hanging system generally uses a two-dimensional monitoring interface to display device state parameters. Once a fault occurs, the operator needs to manually switch between multiple independent interfaces to compare displacement sensor readings, RFID logistics data and visual monitoring pictures, and determine the root cause of the fault by experience. The above fault diagnosis method has the following defects: multi-source data is scattered in independent system interfaces, and manual switching and comparison analysis are required frequently; the dynamic relationship in the three-dimensional space of the system cannot be intuitively presented, and the operator cannot intuitively perceive the spatial position and propagation path of the fault, resulting in low positioning efficiency; fault troubleshooting relies on human experience, and the manual method of checking mechanical components one by one is inefficient. These factors have resulted in long average positioning time and high misjudgment rate of fault diagnosis of the track equipment of the current garment hanging system, which has become a technical bottleneck restricting the digital transformation and upgrading of the garment manufacturing industry. SUMMARY

[0004] Therefore, the present application provides a hanging system equipment fault diagnosis method and system, which uses an intelligent analysis model for fault diagnosis and intuitively displays in a digital twin model, solving the problems of difficult equipment fault detection and low fault positioning efficiency in traditional hanging systems.

[0005] In a first aspect, the present application provides a hanging system equipment fault diagnosis method, comprising:

[0006] constructing a digital twin model of the hanging system based on a three-dimensional modeling tool;

[0007] collecting running data of equipment in the hanging system, and binding the running data with the digital twin model;

[0008] extracting features representing the state of the equipment from the running data, analyzing the features based on an intelligent analysis model, and generating a fault diagnosis result;

[0009] Generate a fault warning signal according to the fault diagnosis result, and display it through the digital twin model.

[0010] In one of the embodiments, the intelligent analysis model includes at least one of a vibration prediction model, a deviation anomaly identification model, and a fault reasoning model.

[0011] In one of the embodiments, the intelligent analysis model includes a vibration prediction model, and the extracting features representing the device state from the operation data and analyzing the features based on the intelligent analysis model to generate the fault diagnosis result includes:

[0012] Obtaining historical vibration data from the operation data, the historical vibration data being collected by an acceleration sensor arranged in the suspension system;

[0013] Extracting time domain features and frequency domain features of the historical vibration data;

[0014] Training an LSTM network using the time domain features and the frequency domain features;

[0015] Using the LSTM network to output a vibration trend prediction value of a future period according to current vibration data;

[0016] Generating a fault diagnosis result when the vibration trend prediction value deviates from a preset threshold.

[0017] In one of the embodiments, the intelligent analysis model includes a deviation anomaly identification model, and the extracting features representing the device state from the operation data and analyzing the features based on the intelligent analysis model to generate the fault diagnosis result includes:

[0018] Obtaining a track deviation, a hook throughput, and a pixel ratio of a conveyor belt congestion area from the operation data, wherein the track deviation is measured by a displacement sensor arranged in the suspension system, the hook throughput is measured by an RFID reader-writer arranged in the suspension system, and the pixel ratio of the conveyor belt congestion area is obtained by measuring a material accumulation amount of a conveyor belt in the suspension system;

[0019] Fusing the track deviation, the hook throughput, and the pixel ratio of the conveyor belt congestion area into multi-dimensional data features;

[0020] Using an isolation forest algorithm to label the multi-dimensional data features to obtain abnormal data points;

[0021] Generating a fault diagnosis result when the abnormal data points meet a preset condition.

[0022] In one of the embodiments, the intelligent analysis model comprises a fault inference model, the extracting the features representing the equipment state in the operation data, the analyzing the features based on the intelligent analysis model, and the generating the fault diagnosis result comprises:

[0023] obtaining the operation data and the features;

[0024] judging the operation data and the features based on preset fault logic judgment rules, and outputting fault causes and confidence levels;

[0025] mapping the operation data and the features to a knowledge graph based on an improved graph neural network model, and outputting fault cause probability ranking; the knowledge graph is constructed based on the fault tree and historical maintenance records; the fault tree is constructed based on fault codes and causal relationships;

[0026] generating a fault diagnosis result according to the fault causes, the confidence levels, and the fault cause probability ranking.

[0027] In one of the embodiments, the method further comprises:

[0028] obtaining newly added maintenance records, the maintenance records comprising fault phenomena, maintenance processes, and root cause analysis;

[0029] performing semantic analysis on the maintenance records by using a natural language processing model to generate edge relationships of the knowledge graph;

[0030] updating the edge relationships to the knowledge graph, and triggering incremental learning of the improved graph neural network model.

[0031] In one of the embodiments, before the extracting the features representing the equipment state, the operation data is preprocessed, and the preprocessing comprises at least one of the following:

[0032] performing wavelet transform processing on the current vibration data and the historical vibration data to eliminate high-frequency interference;

[0033] performing Kalman filter processing on the track offset to compensate for measurement errors of displacement sensors arranged in the suspension system;

[0034] performing timestamp alignment repair on the hook throughput to eliminate clock bias of the RFID reader arranged in the suspension system;

[0035] detecting and counting image data of the conveyor belt state in the suspension system by using a target detection algorithm to obtain a pixel proportion of a congestion area of the conveyor belt.

[0036] In one of the embodiments, the failure warning signal comprises at least one of the following: marking an abnormal device in red in the digital twin model, automatically popping up a list of possible failure causes, generating a standardized maintenance work order, pushing an optimal maintenance time window, and automatically reserving operation logs before and after the failure.

[0037] In one of the embodiments, the generating a failure warning signal according to the failure diagnosis result and displaying the failure warning signal through the digital twin model comprises:

[0038] The failure diagnosis result is classified according to preset failure types, and an importance weight is set for each failure type, wherein the preset failure types comprise at least one of the following: mechanical health, environmental risk, energy consumption indicator, and logistics risk.

[0039] For all failure types identified in the failure diagnosis result, a weighted comprehensive score is calculated according to the weight, and a graded warning signal is generated according to a preset threshold interval to which the weighted comprehensive score belongs.

[0040] In a second aspect, the present application further provides a hoisting system equipment failure diagnosis system, comprising:

[0041] A model construction module is configured to construct a digital twin model of the hoisting system based on a three-dimensional modeling tool.

[0042] A data binding module is configured to collect running data of equipment in the hoisting system and bind the running data to the digital twin model.

[0043] A failure diagnosis module is configured to extract features representing equipment states from the running data, analyze the features based on an intelligent analysis model, and generate a failure diagnosis result.

[0044] An interactive display module is configured to generate a failure warning signal according to the failure diagnosis result and display the failure warning signal through the digital twin model.

[0045] The hoisting system equipment failure diagnosis method and system described above bind equipment running data to a unified digital twin model, eliminating the operation redundancy of manually switching system interfaces; intuitively display equipment failure through a three-dimensional model, improving the positioning efficiency of the failure; extract features based on an intelligent analysis model and automatically generate a failure diagnosis result, reducing the risk of manual experience misjudgment, avoiding inefficient one-by-one component troubleshooting, and shortening the failure positioning time. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application or the related art. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained on the basis of these drawings without creative effort.

[0047] Figure 1 An application environment diagram of the hoisting system equipment fault diagnosis method in an embodiment;

[0048] Figure 2 A flowchart of the hoisting system equipment fault diagnosis method in an embodiment;

[0049] Figure 3 A flowchart of the fault diagnosis method based on a vibration prediction model in an embodiment;

[0050] Figure 4 A flowchart of the fault diagnosis method based on an offset anomaly identification model in an embodiment;

[0051] Figure 5 A flowchart of the fault diagnosis method based on a fault reasoning model in an embodiment;

[0052] Figure 6 A flowchart of the updating method of the knowledge graph for fault reasoning in an embodiment;

[0053] Figure 7 A flowchart of the method for generating a fault warning signal in an embodiment;

[0054] Figure 8 A structural block diagram of the hoisting system equipment fault diagnosis system in an embodiment. DETAILED DESCRIPTION

[0055] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0056] The hoisting system equipment fault diagnosis method provided by the embodiments of the present application can be applied to, for example Figure 1The environment shown. Among them, the physical device layer 10 includes hanging device entities and controllers, sensor clusters, and the physical device layer 10 communicates with the server 11 through the network. The data storage system 12 stores real-time and historical data, event logs, fault rules, and intelligent analysis models and digital twin models, and other data that need to be processed. The data storage system can be integrated on the server 11, or placed on the cloud or other network servers. The server 11 includes an intelligent analysis layer 110 and a digital twin layer 111, which can be used to execute the hanging system device fault diagnosis method in the embodiment. Among them, the intelligent analysis layer 110 obtains the fault diagnosis result by using the data of the data storage system 12 and the state information provided by the digital twin layer 111, and the digital twin layer 111 maps the state information of the physical device layer 10 in real time. The fault diagnosis result obtained by the intelligent analysis layer 110 and the device three-dimensional model in the digital twin layer 111 are displayed to the user through the user terminal 13 application layer. Among them, the user terminal 13 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 11 can be a standalone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0057] In one exemplary embodiment, as Figure 2 shown, a hanging system device fault diagnosis method is provided, as Figure 2 shown, the method comprises the following steps S201 to S204:

[0058] S201, constructing a digital twin model of the hanging system based on a three-dimensional modeling tool.

[0059] The hanging system is an intelligent material transportation system in the garment manufacturing industry that transmits clothes through rails and hooks. It generally includes rail systems, hangers, workstations, driving and control devices, and other core equipment. Among them, the rail system mainly includes the main rail and the auxiliary rail, which are used to support and guide the movement of the hanger; the hanger includes hooks, clamps and other structures, and is the component that directly carries the garment; the workstation provides a stopping device and a working space for the operator. Through the rail, hook, workstation and other equipment, a material transportation network is formed, which can realize the automatic circulation of clothes in each production link.

[0060] The digital twin model technology is a virtual mapping body of a physical system based on three-dimensional modeling technology. By binding real-time sensor data, the physical entity and the virtual model can be synchronized bidirectionally. The digital twin model can dynamically display the device running state, simulate the production process, and make trend prediction based on historical data to support intelligent decision-making. Specifically, professional modeling tools such as BIM, Blender or Unity3D can be used to model the physical components of the hanging system.

[0061] S202, collecting running data of the equipment in the hanging system, and binding the running data with the digital twin model.

[0062] It can be understood that the running data is a data set reflecting the physical state of the equipment in the system at the current and historical time. Specifically, the running data can be collected by multiple types of sensors, including: RFID reader, laser displacement sensor, acceleration sensor, environmental sensor, industrial camera, PLC controller, etc. Among them, the RFID reader is a device used for wireless communication with the RFID tag, which can read the information in the tag through the radio frequency signal, or write data to the writable tag. In this embodiment, RFID is used to count the material throughput of each station, and the unique identifier bound to the clothes is read to realize real-time tracking and counting of material flow. The laser displacement sensor can be used to detect the offset change data of the track and other equipment in the track system. The acceleration sensor is used to collect vibration data of the equipment, and the sampling points include but are not limited to motor bearings, tracks and other components. The environmental sensor detects parameters such as temperature, humidity and dust concentration in the workshop, which can feedback the state of the production environment. The industrial camera can combine computer vision models such as YOLOv8 to identify scenes such as conveyor belt congestion and abnormal material placement, which is convenient for subsequent statistics and output of congestion area pixel ratio and other data. PLC is used to obtain the current, temperature and other equipment running parameters of the motor, which is used as a data source for analyzing the load change of the motor and judging the health status of the equipment.

[0063] Through the preset data mapping rule, the above collected multi-source data is dynamically bound to the corresponding node in the digital twin model, realizing the running state synchronization of the virtual model and the physical equipment.

[0064] S203, extracting features representing the state of the equipment in the running data, analyzing the features based on an intelligent analysis model, and generating a fault diagnosis result.

[0065] The data-driven fault diagnosis method usually needs to preprocess the original data to improve the prediction effect of the model. In this step, different preprocessing methods need to be used for the multi-source data collected in step S202, which can specifically include data filtering, noise reduction, timestamp alignment, etc. The preprocessed multi-source data is extracted by various feature extraction methods to generate a multi-dimensional feature vector, which is input into the intelligent analysis model.

[0066] The intelligent analysis model can be a pre-trained model, which can specifically include algorithm models of machine learning, rule engine, etc., for anomaly detection, trend prediction and root cause reasoning of device features. The fault diagnosis result is a device state evaluation conclusion generated based on feature analysis, which contains fault type and possible cause list, providing basis for subsequent early warning.

[0067] S204, generating a fault warning signal according to the fault diagnosis result and displaying it through the digital twin model.

[0068] The fault diagnosis result can include fault equipment and fault cause, different warning information is generated according to different fault diagnosis results, and different warning levels can be divided according to the severity of the fault. In the three-dimensional picture of the digital twin model, the warning signal is displayed in a visual form, which is convenient for the staff to locate the fault and take corresponding measures.

[0069] In the above hanging system device fault diagnosis method, the digital twin model is used to realize the bidirectional mapping of physical equipment and virtual model, and the fault information of the equipment is displayed through a three-dimensional visual interface, which is convenient for the operator to intuitively view the equipment operation information and locate the fault, avoiding frequent switching of system interfaces. Through intelligent algorithm analysis of multi-source feature data, automatic fault identification is realized, solving the problem of manual data analysis and difficult equipment anomaly positioning in traditional solutions.

[0070] In some embodiments, the running data is preprocessed before the features representing the device state are extracted. It can be understood that different preprocessing methods can be used for different data, and exemplary methods can be taken as follows:

[0071] Wavelet transform is performed on the current vibration data and the historical vibration data to eliminate high-frequency interference.

[0072] Due to the existence of environmental noise and sensor self-noise, high-frequency components often exist in the vibration signal, which will affect the frequency analysis. Wavelet transform can effectively separate high-frequency noise and low-frequency effective signal by decomposing the vibration signal into wavelet basis functions of different frequencies to realize multi-resolution analysis of the signal. Specifically, in the present embodiment, the vibration time-domain signal collected by the three-axis acceleration sensor at a sampling rate of 1 kHz is decomposed into an approximate component and a detail component by using a Daubechies5 wavelet basis for 5-layer decomposition. The threshold value is set for soft threshold processing of the detail component, and after filtering out the high-frequency noise, the signal is reconstructed by inverse wavelet transform to retain the 0-500Hz effective frequency band while eliminating high-frequency interference.

[0073] The track offset is subjected to Kalman filtering to compensate for the measurement error of the displacement sensor arranged in the suspension system.

[0074] The Kalman filter is used to fuse the sensor measurement value and the system state prediction value to obtain the optimal estimation through iterative optimization, which can effectively smooth the noise data and compensate for the measurement error of the laser displacement sensor.

[0075] The timestamp alignment repair is performed on the hook throughput to eliminate the clock deviation of the RFID reader arranged in the suspension system.

[0076] The hook throughput data is measured by the RFID reader. Due to the hardware differences of each RFID device or network delay, the acquisition timestamp may deviate. By calibrating the timestamp, the counting error caused by the asynchronous clock of each device can be eliminated.

[0077] The image data of the conveyor belt state in the suspension system is detected and counted using a target detection algorithm to obtain the pixel ratio of the congested area of the conveyor belt.

[0078] For example, the target detection algorithm can be YOLOv8 algorithm. YOLOv8 is an end-to-end deep learning model. Specifically, the industrial camera is used to collect conveyor belt images at a specified frame rate, and the pre-trained YOLOv8 model is input. The model can detect the material accumulation area in the image and output the bounding box coordinates and confidence. The confidence threshold is set to 0.7. For the bounding box with a confidence greater than 0.7, the ratio of the total number of congested pixels to the total number of pixels in the image is calculated to obtain the congested pixel ratio.

[0079] Through the above preprocessing steps, the noise of the running data is removed, calibrated and standardized, which lays a data foundation for subsequent feature extraction and intelligent analysis, and ensures the accuracy of the fault diagnosis result.

[0080] In some embodiments, the intelligent analysis model in step S203 includes at least one of a vibration prediction model, an offset anomaly recognition model, and a fault reasoning model.

[0081] Specifically, in some embodiments, as shown in Figure 3 The intelligent analysis model includes a vibration prediction model, and a specific method for generating a fault diagnosis result by the vibration prediction model includes steps S301 to S305:

[0082] S301, historical vibration data is obtained from operation data, wherein the historical vibration data is collected by an acceleration sensor arranged in the suspension system.

[0083] The historical vibration data is a time-domain signal of equipment vibration collected by the acceleration sensor, and contains vibration data in normal operation and abnormal state of the equipment, and is used for training the vibration prediction model.

[0084] S302, time-domain features and frequency-domain features of the historical vibration data are extracted.

[0085] In the embodiment, a 60-second sliding window is used for the historical vibration data to extract the frequency-domain features and the time-domain features. The feature extraction method includes but is not limited to: for the time-domain features, RMS value, peak factor (peak value / RMS), and waveform factor (peak value / average value) in each window are calculated; for the frequency-domain features, FFT transformation is performed on the data in each sliding window to extract 1 / 3 octave energy distribution, and energy proportion of bearing fault characteristic frequency and harmonic components is identified. The above time-domain features and frequency-domain features are used to construct a feature vector as an input of an LSTM network model.

[0086] S303, the LSTM network is trained using the time-domain features and the frequency-domain features.

[0087] The long short-term memory network (LSTM) is a recurrent neural network that processes long-term dependent time series data through a gating mechanism, and is suitable for predicting dynamic parameters such as vibration trend. The initialized LSTM network includes an input layer, a hidden layer, and an output layer, and can also include a Dropout layer to prevent overfitting according to specific conditions, define a loss function and an optimizer for network training, and set the number of training rounds and the batch size. The feature vector is divided into a training set and a test set, the LSTM model is trained through the training set, and the loss change and the validation set performance in the training process are monitored until the model meets the expectation.

[0088] S304, the LSTM network is used to output a vibration trend prediction value of a future period according to current vibration data.

[0089] After the LSTM network model is trained, the real-time collected vibration data is extracted according to the same method in step S302 to generate the current feature vector. The current feature vector is input into the trained LSTM model, and the model outputs the vibration prediction value of the next 10 seconds to form a prediction curve.

[0090] S305, when the vibration trend prediction value deviates from the preset threshold, a fault diagnosis result is generated.

[0091] A prediction error threshold is set, and a warning is issued when the actual value deviates from the predicted value by more than the threshold. The threshold can be dynamically adjusted according to business needs, for example, it can be appropriately relaxed during the equipment running-in period.

[0092] In another exemplary embodiment, the fault diagnosis method based on the offset anomaly recognition model is as shown in Figure 4 The intelligent analysis model includes an offset anomaly recognition model, which extracts features representing the state of the device from the operation data, analyzes the features, and generates a fault diagnosis result, which specifically includes steps S401 to S404:

[0093] S401, obtaining the track offset, hook throughput, and pixel ratio of the congested area of the conveying belt from the operation data, wherein the track offset is measured by a displacement sensor arranged in the suspension system, the hook throughput is measured by an RFID reader / writer arranged in the suspension system, and the pixel ratio of the congested area of the conveying belt is obtained by measuring the material accumulation of the conveying belt in the suspension system.

[0094] Specifically, from the data collected by the sensor, the track offset real-time data collected by the laser displacement sensor is extracted, and the time span covers the current production cycle. The hook throughput data recorded by the RFID reader / writer is obtained, the throughput per minute is grouped and counted according to the workstation, and the timestamp is aligned with the track offset data. The image of the conveying belt collected by the industrial camera is called, and the pixel ratio of the congested area of the conveying belt can be generated by detecting the congested area through the YOLOv8 model and calculating the pixel ratio.

[0095] S402, fusing the track offset, hook throughput, and pixel ratio of the congested area of the conveying belt into multi-dimensional data features.

[0096] S403, using the Isolation Forest algorithm to mark abnormal data points for the multi-dimensional data features.

[0097] The Isolation Forest algorithm is an anomaly detection algorithm based on tree structure, which constructs an isolation tree by randomly dividing the feature space. Since the feature values of abnormal points deviate from most samples, the path in the tree is shorter. The path length in the Isolation Forest is the number of path edges from the root node to the leaf node of the data point, and the average path length of the abnormal point is less than that of the normal point.

[0098] First, the isolation forest model is constructed and the number of trees is set. For each three-dimensional feature vector, the path length in each isolated tree is calculated, and the average value is taken as the anomaly score. In this embodiment, by randomly splitting the feature space, the average path length of the abnormal data points in the isolated tree is 8.2, and the normal data points need to pass more than 15 edges to be isolated. The confidence threshold is set to path length < 10, that is, when the path length of the data point is less than 10, it is determined to be an abnormal point.

[0099] Optionally, in the isolation forest algorithm, in order to improve the efficiency of the algorithm, the feature subset can be optimized, that is, the most discriminant feature combination is selected from the high-dimensional features, thereby reducing the data dimension. In this embodiment, since the input feature dimension is high, the Bhattacharyya distance can be used to optimize the feature subset.

[0100] S404, when the abnormal data point meets the preset condition, a fault diagnosis result is generated.

[0101] Since the fault diagnosis result may include multiple faults, in addition to setting the determination rule for each fault, a composite fault rule can also be set, such as when the displacement offset is > 3mm and the congestion pixel ratio is > 30%, it is determined to be a certain type of fault, and the corresponding warning signal is marked and generated. The warning signal includes the warning level, the abnormal feature combination, such as "track offset 5.2mm + congestion pixel ratio 60%", and the possible cause of the fault, such as "track wear causes offset + material accumulation". The warning signal is displayed through the digital twin interface, including marking the abnormal device model, automatically popping up the fault reason list, and pushing to the maintenance terminal.

[0102] In some other exemplary embodiments, as shown in Figure 5 The intelligent analysis model includes a fault reasoning model, and a method for generating a fault diagnosis result includes steps S501 to S506:

[0103] S501, obtaining running data and features.

[0104] The real-time state data of the equipment collected by the sensor includes physical parameters such as track offset, vibration signal, RFID flow, etc. The running data is preprocessed and features are extracted. The specific implementation can follow the above embodiments, and will not be repeated in this embodiment.

[0105] S502, extracting fault codes and causal relationships based on the equipment manual to generate a fault tree.

[0106] Fault tree is a graphical logical model, taking top-level event as root, decomposing into intermediate events and bottom-level events layer by layer, for deducing fault causal relationship. Among them, the top-level event can be device downtime, hook collision and other fault phenomena, the intermediate event includes vibration exceeding threshold, and the bottom-level event is the root cause of the fault. Specifically, a language model is used to parse the fault diagnosis chapter in the device manual, extract various faults and standardize the coding, such as using “F01” to represent bearing wear. And extract the fault phenomenon and the corresponding reason, form a mapping table of phenomenon and reason, and build a fault tree through the mapping table.

[0107] S503, constructing a knowledge graph based on the fault tree and the historical maintenance record.

[0108] In addition to the device manual, actual fault modes not covered by it can also be further supplemented by historical maintenance records. The historical maintenance records are also parsed using a language model to extract fault phenomena, fault causes and disposal methods, which are supplemented to the knowledge graph.

[0109] The knowledge graph is a knowledge base represented by a graph structure, containing entities and relationships. By converting the fault codes, phenomena and causes in the fault tree into entities in the knowledge graph, and converting the causal relationships into directed edges, the correlation strength can be represented by the weight, and the fault causal chain can be intuitively displayed through the graph structure.

[0110] S504, judging the running data and features based on the preset fault logic judgment rule, and outputting the fault cause and confidence.

[0111] For some fault types, expert knowledge in the device field can be converted into executable structured rules. This detection method is also called fault diagnosis based on expert system. For faults with clear source and statistical confidence, operators can match real-time running data and features by establishing rule logic. Exemplarily, a fault rule condition can be represented as:

[0112] IF vibration frequency = 13.2 Hz AND temperature > 80℃ THEN bearing wear (confidence 92%);

[0113] Among them, the vibration frequency and temperature are the typical characteristics of bearing wear defined by experts, and the confidence comes from the matching accuracy rate in expert experience or historical statistics.

[0114] S505, mapping the running data and features to the knowledge graph based on the improved graph neural network model, and outputting the fault cause probability ranking.

[0115] The graph convolutional neural network model can learn the correlation between entities through node features and graph topology. Specifically, a GCN-LSTM model is selected, which is an improved graph convolutional neural network model. The GCN is a graph convolutional neural network used for reasoning fault root causes, and the LSTM is a long short-term memory network used for processing time series data in features. The model is trained with historical maintenance records as labels, so that the accuracy of the model's prediction of faults reaches the expected value. The current running data and features are input into the improved graph neural network model, the activation probability of each fault node in the knowledge graph is calculated, and the top 3 faults and fault root causes with the highest prediction probability are output.

[0116] S506, generating a fault diagnosis result according to the fault cause, the confidence, and the fault cause probability ranking.

[0117] The fault diagnosis conclusion obtained by integrating the rule reasoning and the improved graph neural network reasoning includes the fault type, the possible cause list, and the treatment suggestion.

[0118] In the above embodiments, the knowledge graph for fault reasoning can be continuously updated in the daily operation of the device, as shown in FIG. 6. Figure 6 As shown in FIG. 6, the knowledge graph can be updated according to steps S601 to S603:

[0119] S601, obtaining a newly added maintenance record, the maintenance record including a fault phenomenon, a maintenance process, and a root analysis.

[0120] S602, performing semantic analysis on the maintenance record by using a natural language processing model to generate an edge relationship of the knowledge graph.

[0121] By using a natural language processing model such as BERT, entity recognition and relationship extraction are performed on the maintenance record text, and the unstructured text is converted into a directed edge relationship such as "vibration anomaly → electromagnetic valve jam". The manual maintenance experience is converted into a structured association of the knowledge graph, solving the problem of lagging knowledge update of the traditional expert system.

[0122] S603, updating the edge relationship to the knowledge graph, and triggering incremental learning of the improved graph neural network model.

[0123] Exemplarily, the newly generated edge relationship such as "vibration anomaly → electromagnetic valve jam" is added to the graph structure of the knowledge graph, and the graph neural network model is incrementally trained only for the newly added knowledge. In this way, new fault modes can be quickly absorbed on the basis of maintaining the original knowledge of the model, so that the diagnosis model continuously adapts to the changes in the operation state of the device, and the recognition ability for new faults is improved.

[0124] In some embodiments, a fault warning signal is generated according to the fault diagnosis result in the above embodiments, and is displayed through the digital twin model, as shown in FIG. 7.Figure 7 As shown, specifically includes step S701 to step S702:

[0125] S701, according to the preset fault type classification of fault diagnosis results, and set the importance weight for each fault type, the preset fault type includes: mechanical health, environmental risk, energy consumption index, logistics risk.

[0126] First, according to the preset fault type, the fault diagnosis result is divided into different categories, including but not limited to mechanical health, environmental risk, energy consumption index, logistics risk four categories. Among them, the mechanical health class fault covers bearing wear, track deviation and other equipment body abnormalities; Environmental risk involves temperature and humidity exceeding the standard, dust concentration is too high and other factors affecting production safety; Energy consumption index focuses on motor current anomaly, energy waste and other efficiency problems; Logistics risk includes hook blockage, conveyor belt congestion and other material transmission failures.

[0127] The single fault type score in the fault diagnosis result is the basis of the weighted comprehensive score, and the single fault type score can be generated by the output result of the corresponding fault diagnosis model. For example, in the fault diagnosis result of the LSTM vibration prediction model, the degree of deviation of the predicted vibration curve from the normal value is quantified as a score. For example, it can be set that more than 15% of the preset value corresponds to 50 points, more than 30% corresponds to 70 points, and so on.

[0128] Set the importance weight for each fault type, for example, mechanical health directly affects the safety of equipment operation, and is given a weight of 40%; Environmental risk is related to production compliance and personnel safety, and the weight is set to 30%; Energy consumption index and logistics risk are given a weight of 20% and 10% respectively, so as to reflect the difference in the influence degree of different faults on the production system.

[0129] S702, according to the importance weight, calculate the weighted comprehensive score of all fault types identified in the fault diagnosis result, according to the preset threshold interval of the weighted comprehensive score, generate a graded warning signal.

[0130] If the diagnosis result contains a mechanical health fault score of 80 points and a logistics risk fault score of 60 points, the weighted comprehensive score is 80x40%+60x10%=38 points. By this calculation method, the influence degree of multiple fault types is quantified as a single numerical value, which is convenient for unified evaluation of the severity level of fault.

[0131] According to the preset threshold interval to which the weighted comprehensive score belongs, a graded early warning signal is generated. For example, the preset threshold interval is: 0-30 points for the warning level, prompting the operator to pay attention to observation; 31-60 points for the deceleration level, automatically triggering the equipment to run at a reduced speed to avoid the expansion of the fault; and 61 points and above for the shutdown level, forcing the equipment to stop running to ensure safety. This graded early warning mechanism can achieve differentiated response to faults, avoiding excessive shutdown caused by minor faults, and taking timely measures before serious faults occur, balancing production continuity and equipment safety.

[0132] In the above embodiments, the early warning signal generated by the fault diagnosis result can include the following: the digital twin interface marks the abnormal equipment in red, automatically pops up a list of possible causes of the fault, displays the congestion propagation path in a three-dimensional perspective, generates a standardized maintenance work order, pushes the optimal maintenance time window, and automatically retains the operation logs before and after the fault.

[0133] Specifically, the digital twin interface marking the abnormal equipment in red can specifically render the abnormal components in red in the three-dimensional visualization scene. The automatically popped-up list of possible causes of the fault is generated based on the causal relationship of the knowledge graph and the probability ranking of the graph neural network. The standardized maintenance work order includes the fault equipment ID, possible causes, recommended maintenance steps, and spare parts list. The optimal maintenance time window with the least impact on production capacity is pushed, the interference of maintenance on production is reduced through intelligent scheduling, and the operation logs 10 minutes before and after the fault are automatically retained, including sensor data curves, operator instruction records, and equipment action trajectories. The log data is used for subsequent fault review and model optimization.

[0134] It should be understood that although each step in the flowchart involved in the above embodiments is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, as described above, at least part of the steps in the flowchart involved in each embodiment can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0135] In an example embodiment, based on the same inventive concept, a hoisting system equipment fault diagnosis system 800 is provided, comprising: a model construction module 801, a data binding module 802, a fault diagnosis module 803, and an interactive display module 804. The implementation scheme for solving problems provided by the fault diagnosis system is similar to the implementation scheme described in the above method, and therefore specific limitations can be referred to the above limitations for the hoisting system equipment fault diagnosis method, which will not be described here again. As shown in Figure 8 The system comprises:

[0136] The model construction module 801 is configured to construct a digital twin model of the hoisting system based on a three-dimensional modeling tool.

[0137] The data binding module 802 is configured to collect operation data of equipment in the hoisting system, and bind the operation data with the digital twin model.

[0138] The fault diagnosis module 803 is configured to extract features representing the state of the equipment from the operation data, analyze the features based on an intelligent analysis model, and generate a fault diagnosis result.

[0139] The interactive display module 804 is configured to generate a fault warning signal according to the fault diagnosis result, and display the fault warning signal through the digital twin model.

[0140] Each module in the hoisting system equipment fault diagnosis system described above can be realized by software, hardware, and combinations thereof, in whole or in part. Each module described above can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations corresponding to each module.

[0141] Those skilled in the art can understand that Figure 8 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0142] Each technical feature of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope recorded in the present application.

[0143] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method of diagnosing a failure of a hoisting system device, characterized by, The method comprises: constructing a digital twin model of the suspension system based on a three-dimensional modeling tool; collecting operation data of equipment in the suspension system, and binding the operation data with the digital twin model; extracting features representing the state of the equipment from the operation data, analyzing the features based on an intelligent analysis model, and generating a fault diagnosis result; generating a fault warning signal according to the fault diagnosis result and displaying it through the digital twin model.

2. The method of claim 1, wherein, The intelligent analysis model includes at least one of a vibration prediction model, an offset anomaly identification model, and a fault reasoning model.

3. The method of claim 2, wherein, The intelligent analysis model includes a vibration prediction model, and the extraction of features representing the state of the equipment from the operation data, the analysis of the features based on the intelligent analysis model, and the generation of the fault diagnosis result comprise: obtaining historical vibration data from the operation data, the historical vibration data being collected by an acceleration sensor arranged in the suspension system; extracting time domain features and frequency domain features of the historical vibration data; training an LSTM network using the time domain features and the frequency domain features; using the LSTM network to output a vibration trend prediction value for a future period according to current vibration data; when the vibration trend prediction value for the future period deviates from a preset threshold, generating a fault diagnosis result.

4. The method of claim 2, wherein, The intelligent analysis model includes an offset anomaly identification model, and the extraction of features representing the state of the equipment from the operation data, the analysis of the features based on the intelligent analysis model, and the generation of the fault diagnosis result comprise: obtaining track offset, hook throughput, and pixel ratio of congested area of the conveyor belt from the operation data, wherein the track offset is measured by a displacement sensor arranged in the suspension system, the hook throughput is measured by an RFID reader / writer arranged in the suspension system, and the pixel ratio of the congested area of the conveyor belt is obtained by measuring the material accumulation of the conveyor belt in the suspension system; fusing the track offset, the hook throughput, and the pixel ratio of the congested area of the conveyor belt into multi-dimensional data features; using an isolation forest algorithm to label the multi-dimensional data features to obtain abnormal data points; when the abnormal data points meet a preset condition, generating a fault diagnosis result.

5. The method of claim 2, wherein, The intelligent analysis model includes a fault reasoning model, and the extraction of features representing the state of the equipment from the operation data, the analysis of the features based on the intelligent analysis model, and the generation of the fault diagnosis result comprise: obtaining the operation data and the features; judging the operation data and the features based on a preset fault logic judgment rule, and outputting a fault cause and a confidence level; mapping the operation data and the features to a knowledge graph based on an improved graph neural network model, and outputting a fault cause probability ranking; the knowledge graph is constructed based on a fault tree and historical maintenance records; the fault tree is constructed based on fault codes and causal relationships; generating a fault diagnosis result according to the fault cause, the confidence level, and the fault cause probability ranking.

6. The method of claim 5, wherein, The method further comprises: obtaining an added maintenance record, the maintenance record including a fault phenomenon, a maintenance process, and a root cause analysis; adopting a natural language processing model to perform semantic analysis on the maintenance record to generate an edge relationship of the knowledge graph; updating the edge relationship to the knowledge graph, and triggering incremental learning of the improved graph neural network model.

7. The method of claim 1, wherein, Before extracting the features representing the device state, the running data is preprocessed, and the preprocessing includes at least one of the following: wavelet transform processing is performed on the current vibration data and the historical vibration data to eliminate high-frequency interference; Kalman filter processing is performed on the track offset to compensate for the measurement error of the displacement sensor arranged in the suspension system; time stamp alignment repair is performed on the hook throughput to eliminate the clock deviation of the RFID reader arranged in the suspension system; target detection algorithm is used to detect and count the image data of the conveyor belt state in the suspension system to obtain the pixel proportion of the congested area of the conveyor belt.

8. The method of claim 1, wherein, The fault warning signal includes at least one of the following: in the digital twin model, the abnormal device is marked in red, the fault possible reason list is automatically popped up, the standardized maintenance work order is generated, the optimal maintenance time window is pushed, and the operation log before and after the fault is automatically reserved.

9. The method of claim 1, wherein, The fault warning signal is generated according to the fault diagnosis result, and the digital twin model is used for display, including: classifying the fault diagnosis result according to a preset fault type, and setting a weight for each fault type, the preset fault type including at least one of mechanical health, environmental risk, energy consumption index and logistics risk; for all fault types identified in the fault diagnosis result, a weighted comprehensive score is calculated according to the weight, and a graded warning signal is generated according to the preset threshold interval to which the weighted comprehensive score belongs.

10. A suspension system equipment failure diagnosis system characterized by comprising: including: a model construction module for constructing a digital twin model of the suspension system based on a three-dimensional modeling tool; a data binding module for collecting running data of devices in the suspension system and binding the running data with the digital twin model; a fault diagnosis module for extracting features representing the device state from the running data, analyzing the features based on an intelligent analysis model, and generating a fault diagnosis result; an interactive display module for generating a fault warning signal according to the fault diagnosis result and displaying the fault warning signal through the digital twin model.

Images