Equipment dynamic management system and method based on real-time data acquisition and intelligent analysis

Abstract

The invention provides an equipment dynamic management system and method based on real-time data acquisition and intelligent analysis, and belongs to the field of high-speed wire harness assembling.The method comprises the steps that real-time data of a high-speed wire harness assembly production line is acquired and preprocessed; inputting the trained edge self-supervision model, and generating a single-device health state code and a collaborative risk feature corresponding to the real-time data; according to the single-device health state codes and the collaborative risk features, completing construction of a dynamic collaborative graph oriented to the high-speed wire harness production line on a cloud platform; performing convolution operation in combination with a joint graph convolution network to obtain a global health score and an abnormal diffusion risk score; and analyzing the global health score and the abnormal diffusion risk score, and generating a grading strategy in combination with the position of the equipment in the production line, the current production takt and the optimization priority of the equipment. The intelligent level of equipment management and the production efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of high-speed wire harness assembly, and in particular relates to an equipment dynamic management system and method based on real-time data acquisition and intelligent analysis. Background Art

[0002] In modern manufacturing, especially in the high-speed wire harness assembly industry, production automation is increasing, and the continuity, stability, and efficiency of production lines have become important indicators of production performance. High-speed wire routing and harness assembly, as key processes in this industry, involve the coordinated operation of multiple highly automated machines, such as wire cutters, strippers, arranging machines, and welding equipment. The operating status of these machines directly impacts product yield, production efficiency, and maintenance costs. However, the current equipment management methods commonly used in the industry still have limitations. Existing technologies often rely on independent health management of individual devices and simple threshold alarm mechanisms. These typically involve installing sensors for temperature, vibration, pressure, and other functions on the equipment to collect operational data and analyze this data using cloud-based servers. However, this approach has significant shortcomings in the complex environment of high-speed wire harness production lines, where multiple devices collaborate.

[0003] First, existing methods lack the ability to comprehensively model and analyze the collaborative relationships between multiple devices. They typically only monitor abnormal signals from a single device and fail to effectively capture potential collaborative risks between devices. For example, in actual production, abnormal vibrations from a wire traversing machine can affect the accuracy of downstream welding equipment, leading to batch product defects. However, this cross-device risk chain cannot be effectively identified using a single-point anomaly detection mechanism.

[0004] Secondly, the data characteristics of the production site are characterized by high frequency and real-time. Especially under high-speed production rhythms, the amount of data generated by the equipment is large and the update speed is fast. The existing "simple collection on the device side + centralized analysis in the cloud" solution is easily limited by bandwidth and cloud computing resources, resulting in insufficient real-time performance and the inability to complete efficient abnormal response at critical moments, thus affecting production continuity and product quality. In addition, some early abnormal signals have hidden and weak signal characteristics, which are difficult to accurately detect through traditional supervised learning models. The reason is that there is a lack of sufficient labeled data for such equipment operation anomalies. Traditional models that rely on a large number of labeled samples are difficult to deploy efficiently in this industry.

[0005] To address these issues, the industry has attempted to introduce some intelligent management methods based on the Internet of Things (IoT) and machine learning. However, most remain at the level of "single-device intelligence" or "lagging cloud-based analysis," failing to develop an intelligent system that integrates end-to-end cloud computing and collaborates locally and globally. This makes it difficult to balance real-time performance, collaboration, and predictive accuracy. Therefore, establishing a system that addresses the unique characteristics of the high-speed wire harness assembly industry, enables real-time response, multi-device collaboration, and possesses intelligent prediction and dynamic management capabilities has become a key issue that the industry urgently needs to address. Summary of the Invention

[0006] This invention aims to design a dynamic device management system and method based on real-time data collection and intelligent analysis. By designing a two-tier "end-cloud collaborative" architecture, leveraging the intelligent analysis capabilities of edge devices and combining them with the global health prediction mechanism of a central platform, this approach overcomes the technical bottlenecks of traditional solutions, characterized by "single-point monitoring" and "delayed response."

[0007] In order to achieve the above-mentioned object, in a first aspect of the present invention, a method for dynamic management of equipment based on real-time data collection and intelligent analysis is provided, the method comprising:

[0008] Acquire real-time data from high-speed wire harness assembly lines and perform pre-processing;

[0009] Inputting the preprocessed data into a trained edge self-supervisory model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data;

[0010] Based on the single-device health status coding and collaborative risk characteristics, a dynamic collaborative graph for a high-speed wire harness production line is constructed on a cloud platform;

[0011] Based on the dynamic collaborative graph, combined with the joint graph convolutional network, a convolution operation is performed to obtain a global health score and anomaly diffusion risk score. The Euclidean distance analysis between embedded features is performed based on the diffusion risk score to achieve highly sensitive prediction of anomaly diffusion chains.

[0012] The global health score and anomaly spread risk score are analyzed, and a grading strategy is generated based on the device's position in the production line, the current production rhythm, and the device's optimization priority. The system then performs anomaly analysis operations in conjunction with the dynamic collaborative graph. When a device is identified as a high-priority anomaly source, the system automatically selects associated downstream devices based on their link position in the dynamic collaborative graph to synchronously execute mitigation actions.

[0013] The pre-processed data includes a weighted fusion feature vector of the current period of the device and a sensitivity factor of the current feature fluctuation of the device;

[0014] The trained edge self-supervision model includes at least an anomaly sensitivity modulation term and a collaborative regularization term. The anomaly sensitivity modulation term is used to control the degree of improvement of the anomaly separation capability of high-risk equipment; the collaborative regularization term is used to measure the degree of deviation between the characteristics of the equipment and the historical collaborative fault graph of the downstream equipment.

[0015] Preferably, the real-time data includes the production rhythm, load status and historical abnormality rate of the current equipment, as well as the vibration, temperature and pressure signals of the equipment;

[0016] Wherein, the preprocessing includes:

[0017] Dynamically adjust the time window lengths of different signals according to the production rhythm, load status, and historical abnormality rate of the current equipment;

[0018] Obtaining a weighted fusion feature vector of the current period of the device by weighted fusion according to the vibration, temperature, and pressure signals of the device;

[0019] A dynamic weighted standard analysis is performed based on the weighted fusion feature vector of the current period of the device to obtain a sensitivity factor of the current feature fluctuation of the device for abnormality warning.

[0020] Preferably, the trained edge self-supervisory model adopts an improved contrastive learning framework and is deployed in the edge device.

[0021] Preferably, the trained edge self-supervision model has a loss function Expressed as:

[0022]

[0023] in, Training target for edge self-supervised model; z i Encode the health status of device i, which represents the potential health distribution of the device; For device i, feature F i A weakly enhanced version of is the abnormal sensitivity factor of device i; β is the sensitivity modulation factor, which controls the degree of improvement of the abnormal separation capability of high-risk devices; is the collaborative regularization term, which measures the characteristics of device i and the historical collaborative failure graph G of downstream devices hist The degree of deviation; η is the collaborative regularization weight;

[0024] The collaborative regularization term is related to the distance between the health status code of the device and the historical collaborative anomaly center.

[0025] Preferably, the structure of the dynamic collaboration graph is:

[0026] Node set V = {v i}: Each v i Represents production line equipment i, and the node feature is the health status code z i ;

[0027] Edge set E={e i,j}: The synergistic edge between equipment i and equipment j. The strength of the edge is affected by the production process dependency and the synergistic risk characteristic C of equipment i. i , Equipment j Collaborative Risk Characteristics C j Dynamic impact;

[0028] Collaborative risk set C = {C i}: Synchronously obtain from the edge side;

[0029] Among them, the collaborative edge weight e i,j The dynamic generation of is completed by the following formula:

[0030]

[0031] Among them, e i,j is the edge weight between device i and device j; κ i,j is the static process dependency value in the production process; δ is the adjustment coefficient, which is set for the global production rhythm; C i 、C j The collaborative risk signature generated for the edge indicates the potential for device anomaly propagation.

[0032] Preferably, the convolution operation of the joint graph convolutional network is specifically:

[0033] Use node features as health status encoding z i as the node embedding feature of device i at layer 0;

[0034] The sensitivity factor of the current characteristic fluctuation of the device is used as the diffusion modulation item;

[0035] For the node embedding features of device i at layer 0, that is, node i, then aggregate the features of its neighbor node j, weight the neighbor features with the weight matrix, and introduce a diffusion modulation term to enhance or suppress the influence of neighbor information. Finally, the updated node features are output through the activation function σ; the updated node features include the global health score of the device and the abnormal diffusion risk score of the device.

[0036] Preferably, the joint graph convolutional network also includes a collaborative propagation regularization term, which is used to output the abnormal diffusion risk score of the device by analyzing the Euclidean distance between the final embedded features of device i and neighbor j; wherein the Euclidean distance reflects the difference between the two in the feature space, and the largest difference means the highest propagation potential.

[0037] Preferably, the global health score and the abnormal diffusion risk score are analyzed, and three grading strategies are generated in combination with the position of the equipment in the production line and the current production rhythm: a high priority strategy, a medium priority strategy and a low priority strategy.

[0038] Preferably, after obtaining the three classification strategies, the method further includes:

[0039] Obtain the global health score and abnormal spread risk score, and perform multi-factor weighted analysis to obtain the optimization priority score of the equipment;

[0040] In combination with the optimization priority score of the device, a specific control strategy is generated; the specific control strategy includes:

[0041] When the optimization priority score is greater than the highest threshold, an emergency mitigation strategy is issued to coordinate optimization of the device and its downstream devices, and joint measures are taken.

[0042] When the optimization priority score is less than the highest threshold but greater than the middle threshold, a single-point self-healing policy is issued to perform local control optimization only on device i.

[0043] When the optimization priority score is lower than the intermediate threshold, only a monitoring report is generated, indicating that the device is healthy but requires further observation, and the control policy is not triggered.

[0044] After receiving the control strategy, the edge performs lightweight strategy execution operations, including: parsing control instructions; executing the strategy in real time and feeding back the execution results to the cloud; ensuring that the strategy is quickly implemented under the high-speed operation of the production line, maintaining the continuity and safety of the production line.

[0045] In a second aspect of the present invention, there is provided a device dynamic management system based on real-time data collection and intelligent analysis, the system comprising:

[0046] Production line data acquisition module, used to obtain real-time data of high-speed wire harness assembly production line and perform pre-processing;

[0047] An edge self-supervision module is used to input the pre-processed data into a trained edge self-supervision model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data;

[0048] A cloud-based collaborative graph module is used to construct a dynamic collaborative graph for a high-speed wire harness production line on a cloud platform based on the health status code of the single device and the collaborative risk characteristics;

[0049] A GNN prediction module is used to perform convolution operations based on the dynamic collaborative graph and a joint graph convolutional network to obtain a global health score and anomaly diffusion risk score, and perform Euclidean distance analysis between embedded features based on the diffusion risk score to achieve highly sensitive prediction of anomaly diffusion chains;

[0050] An optimization execution module is used to analyze the global health score and anomaly diffusion risk score, and generate a classification strategy based on the device's position in the production line, the current production rhythm, and the device's optimization priority, and perform anomaly analysis operations in conjunction with the dynamic collaboration graph. When a device is determined to be a high-priority anomaly source, the system automatically selects associated downstream devices based on their link position in the dynamic collaboration graph to synchronously execute mitigation actions.

[0051] The pre-processed data includes a weighted fusion feature vector of the current period of the device and a sensitivity factor of the current feature fluctuation of the device;

[0052] The trained edge self-supervision model includes at least an anomaly sensitivity modulation term and a collaborative regularization term. The anomaly sensitivity modulation term is used to control the degree of improvement in the anomaly separation capability of high-risk equipment; the collaborative regularization term measures the degree of deviation between the characteristics of the equipment and the historical collaborative fault graph of the downstream equipment.

[0053] The beneficial technical effects of the present invention are at least as follows:

[0054] In response to the above-mentioned deficiencies in equipment management of existing high-speed wire harness assembly production lines, the present invention proposes a dynamic equipment management system based on real-time data collection and intelligent analysis, which can realize real-time health management and dynamic optimization of multiple collaborative equipment. The innovation of the present invention lies in the design of a two-tier architecture of "end-cloud collaboration", which utilizes the intelligent analysis capabilities of the edge device side and combines the global health prediction mechanism of the central platform to break through the technical bottlenecks of "single-point monitoring" and "response lag" of traditional solutions. By implementing lightweight real-time anomaly detection on the device side, it can quickly respond to equipment anomalies and ensure real-time requirements under high-speed production. At the same time, the global analysis module integrated in the system can dynamically predict the potential chain failure risks between equipment groups based on the collaborative relationship of multiple devices, solving the problem that existing solutions cannot effectively deal with collaborative anomalies.

[0055] Furthermore, this invention innovatively addresses the model training challenge caused by a scarcity of abnormal samples. It designs a self-learning mechanism based on equipment operating data, improving the system's ability to identify early, hidden anomalies, thereby effectively reducing equipment failure rates and the risk of production interruptions. This overall solution not only enables real-time equipment status monitoring and anomaly detection, but also dynamically adjusts maintenance strategies and optimizes production scheduling, ensuring the continuity and efficiency of production lines and significantly improving the intelligence level of equipment management and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0057] Figure 1 This is a flow chart of a device dynamic management method based on real-time data collection and intelligent analysis of the present invention.

[0058] Figure 2 This is a framework diagram of an equipment dynamic management system based on real-time data collection and intelligent analysis in the present invention. DETAILED DESCRIPTION

[0059] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0060] In one or more embodiments, Figure 1 As shown, a device dynamic management method based on real-time data collection and intelligent analysis is disclosed, and the method includes the following steps S1-S5:

[0061] S1. Acquire real-time data from a high-speed wire harness assembly line and perform preprocessing.

[0062] Specifically, on the high-speed wire harness assembly production line, multiple collaborative equipment (such as wire arranging machines, wire cutting machines, welding machines, etc.) are involved. Due to the different execution processes and complex physical structures of each device, its operating state has significant non-uniform characteristics of "high-frequency vibration signal bursts, slow changes in temperature signals, and periodic fluctuations in pressure signals". These devices work under high load and high speed (the production cycle is in seconds or even sub-seconds). The traditional "unified time window + fixed weighted feature fusion" solution cannot accurately capture the multi-scale characteristics of equipment anomalies, and it is difficult to serve subsequent intelligent prediction and dynamic management. Therefore, this step proposes a "dynamic working condition adaptive fusion coding mechanism" specifically for high-speed wire harness production conditions, which combines the time windows, weighting mechanisms, feature generation of different signals with the real-time working conditions of the equipment to form a set of industry-customized "intelligent perception and coding" systems.

[0063] First, based on the real-time working condition of the equipment C i (including the production rhythm r of the current equipment i i , load status i and historical anomaly rate e i), dynamically adjust the time window length of different signals to solve the problem of inconsistent time scales of signals across modalities. For example, when the speed of the wire traversing machine is r i When it reaches high speed area (greater than 2000rpm), the vibration signal The time window is automatically shortened to 0.1 seconds (original design is 1 second), and the temperature signal The window is kept for 1 second, while the pressure signal According to l i Decide whether to shorten it to 0.5 seconds, forming a "multi-scale time window mechanism that can be scaled on demand" to dynamically adapt to the current production rhythm.

[0064] The collected multimodal signals (k=1,2,3 represent vibration, temperature and pressure signals respectively) After standardization and denoising, an "industry-specific dynamic weighted fusion model" is designed. i and the contribution of historical abnormal events, giving each signal channel a dynamic weight And for the first time, the "abnormal correlation factor" was introduced into the weighted model. ”, this factor reflects the synergy strength between signal k and historical anomalies of device i.

[0065]

[0066] Among them, F i Represents the weighted fusion feature vector of the current cycle of device i, which is used for subsequent anomaly detection; Indicates the dynamic weighting coefficient of signal channel k, based on the equipment working condition C i Dynamic adjustment; The correlation factor between signal k and the historical anomaly of the device, reflecting the contribution to the anomaly; It represents the value after feature extraction (such as RMS, peak value, etc.) of signal channel k; K represents the number of signal channels (3 in this example).

[0067] For example, when the wire arrangement machine is processing at high speed, the vibration signal is given a higher Temperature signal Pressure signal Highlight the criticality of vibration signals in anomaly detection.

[0068] In order to further improve the system's collaborative perception of "early weak anomalies" and "strong sudden anomalies", a "fluctuation sensitivity adjustment mechanism" combined with dynamic weighting is proposed to generate anomaly sensitivity characteristic factors. To measure F i The difference between the volatility of the historical steady-state volatility level and the innovative addition of the “co-regularization term” λ i , reflecting the risk of coordinated failure between device i and upstream and downstream devices.

[0069]

[0070] in, Indicates the sensitivity factor of the current characteristic fluctuation of device i, which is used for abnormal warning; std(F i ) indicates that F i The standard deviation reflects the signal fluctuation; Indicates that the device is in normal historical state. i The average value; ∈ represents a small constant to prevent division by zero (such as 10 -6 );λ i It represents the collaborative anomaly risk factor between device i and other devices, reflecting the possibility of anomaly propagation.

[0071] For example, there is a resonance relationship between the wire traversing machine and the downstream welding equipment, λ i It can be set to 0.3 to 0.5 to make the system more sensitive to abnormal fluctuations at the workstation.

[0072] The final output of this step is F i and It will be used as the direct input of the next step (edge ​​self-supervision model), F i As the local health feature input model M SSL , As a collaborative sensitivity prior, it participates in the modeling process of abnormal features. This design enables edge intelligent analysis to not only focus on single-point feature fluctuations but also possess a certain degree of "collaborative perception capability," providing highly sensitive, collaboratively perceived basic data for subsequent collaborative feature generation and global modeling in the cloud.

[0073] S2. Input the preprocessed data into the trained edge self-supervision model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data.

[0074] Specifically, this step is based on the output result of step 1, and receives the fused feature vector F from the device i in the high-speed wire harness assembly line. i and abnormal sensitivity factor The "collaborative feature generation of edge self-supervision models" function is implemented at the edge. This step aims to solve the following core problems: high-speed wire harness production lines have complex multi-device collaboration and anomalies are propagating, while traditional edge anomaly detection models can only identify the health status of a single device and ignore the collaborative risk propagation characteristics of anomalies, resulting in insufficient early warning capabilities for "chain failures." To address the above pain points, this step innovatively designs a "collaborative sensitive self-supervision model" M SSL , not only completes the local health status coding of device i z iAt the same time, for the "collaborative risk propagation effect" of high-speed wire harness industry equipment, the risk feature C with collaborative perception capability is output. i Model M SSL F generated in step 1 i and With the information as input, the self-supervised learning mechanism is used to realize the joint modeling of equipment’s own operation mode learning, abnormal potential discrimination and collaborative propagation factors.

[0075] To achieve this goal, M SSL By adopting an improved contrastive learning framework and innovatively introducing the "abnormal sensitivity modulation term" and "cooperative regularization term", a more practical training objective function is designed for the multi-device collaborative working conditions in the high-speed wire harness industry. The specific self-supervised training loss is:

[0076]

[0077] in, represents the edge self-supervised model training target; z i The health status code of device i represents the potential health distribution of the device; Indicates the device i feature F i A weakly enhanced version of represents the abnormal sensitivity factor of device i; β represents the sensitivity modulation factor, which controls the degree of improvement in the abnormal separation capability of high-risk devices; represents the collaborative regularization term, which measures the characteristics of device i and the historical collaborative failure graph G of downstream devices hist The degree of deviation; η represents the collaborative regularization weight.

[0078] Among them, the collaborative regularization term It is an innovative regular constraint designed in combination with the high-speed wiring harness industry collaborative relationship diagram. Its calculation method is based on the state code z of device i. i Collaborative Anomaly Center with History distance, while considering the edge coordination factor λ between device i and downstream device j i,j :

[0079]

[0080] Where N(i) represents the set of downstream directly adjacent devices of device i; i,j Indicates the cooperative transmission intensity coefficient of equipment i and equipment j in the historical anomaly graph (for example, there is a high resonance risk between the wire traversing machine and the downstream welding equipment, λ i,j Can be set to 0.5); Represents the historical collaborative anomaly center vector of downstream device j.

[0081] By introducing collaborative regularization on the edge side, the model not only focuses on the health representation of the device (z i ), and its synergistic risk potential C i Explicit modeling is performed, As C i This core computational basis is directly used in subsequent cloud-based collaborative graph modeling (G(V, E, C)), forming an "edge-to-cloud collaborative feature chain." This design, for the first time, connects "edge health" with "inter-device collaborative transmission risk," breaking through the traditional design bottleneck of "edge anomaly detection focusing solely on local areas."

[0082] After the model is deployed, in the actual production process, the edge device receives F i and Execute M SSL Reasoning, generating two types of output:

[0083] z i : Single device health status encoding as subsequent GNN graph node features;

[0084] C i : Synergistic risk characteristics, defined as Reflects the potential impact of device i anomalies on downstream devices.

[0085] For example, in actual scenarios, when the wire arrangement machine is slightly worn due to bearing Increased to 1.6, M SSL The training mechanism will improve its ability to distinguish abnormalities (affected by At the same time, due to the synergy coefficient λ between the wire arrangement machine and the downstream welding machine i,j =0.5, the final calculated C i =0.5, the subsequent steps in the cloud will be based on C i In the collaborative graph, the collaborative risk of the edge between the wire arranging machine and the welding machine is significantly increased, reminding the cloud-side GNN to focus on the potential risk spread of this link.

[0086] S3. Based on the single device health status coding and collaborative risk characteristics, a dynamic collaborative graph for the high-speed wire harness production line is constructed on the cloud platform.

[0087] Specifically, the core goal of this step is to use the edge self-supervision model M in step 2 SSL Generated device health status code z i and synergistic risk signature C i, complete the construction of the dynamic collaborative graph G(V, E, C) for the high-speed wire harness production line on the cloud platform as the input for the subsequent collaborative health prediction (step 4). Combined with the characteristics of the high-speed wire harness assembly industry, there is a strong collaborative dependency between the various equipment in the production line (such as wire arrangement machines, wire cutting machines, welding machines, and testing machines), and the physical coupling between equipment, abnormal resonance propagation, beat chain and other characteristics are significant. Therefore, the traditional "static process flow chart" is difficult to reflect the dynamic changes of the collaborative relationship in real time. In response to the above problems, the present invention proposes a "dynamic mapping mechanism driven by collaborative risk", which relies on C from the edge side in the cloud. i (Collaborative Risk Characteristics) Dynamically construct graph structures to address the industry pain point of the lack of real-time topology of the collaborative graph of equipment on the production line.

[0088] This step mainly completes the generation of the graph structure G(V, E, C), which specifically includes the following key subtasks:

[0089] Node set V = {v i}: Each v i Represents production line equipment i, and the node feature is the health status code z generated in step 2 i ;

[0090] Edge set E={e i,j}: The synergistic edge between equipment i and equipment j, the strength of which is affected by the production process and C i 、C j Dynamic impact;

[0091] Collaborative risk set C = {C i}: Acquired synchronously from the edge side as the input of the subsequent graph model.

[0092] Collaborative edge weight i,j The dynamic generation of is accomplished by the following “synergistic risk driving formula”:

[0093]

[0094] Among them, e i,j represents the edge weight between device i and device j; κ i,j Represents the static process dependency value in the production process (such as from the wire arrangement machine to the wire cutting machine, κ i,j =0.6); δ represents the adjustment coefficient, which is set for the global production rhythm (e.g., δ = 0.5); C i 、C j It represents the collaborative risk characteristics generated at the edge and indicates the potential for device anomaly propagation.

[0095] Through the above mechanism, e i,j No longer relying solely on traditional process structures, but real-time perception of the collaborative risk level of equipment. For example, when a local abnormality (Ci =0.6) and the downstream wire cutting machine collaborative risk C j = 0.5, the edge weight is dynamically increased, truly reflecting the current abnormal propagation trend. The resulting collaborative graph G(V, E, C) has the following characteristics:

[0096] Node v i :Contains z i , directly used as the node feature of the subsequent GNN model;

[0097] Edge i,j : Dynamic edge weights quantify the intensity of anomaly propagation between devices;

[0098] Graph topology: reflects the physical dependencies and current collaborative status of the production line.

[0099] For example, when the production cycle is increased (δ = 0.5), the history of the wire arranging machine and the welding machine κ i,j =0.7, if C i =0.6, C j =0.5, then the edge weight is:

[0100] e i,j =0.7·(1+0.5·0.55)=0.7·1.275=0.8925.

[0101] The dynamic increase in edge weight reflects the enhanced propagation effect of abnormalities in the wire arranging machine on the risks of the welding machine under the current high-beat condition.

[0102] This step finally outputs G(V,E,C) as the input of step 4.

[0103] S4. Based on the dynamic collaborative graph, combined with the joint graph convolutional network, a convolution operation is performed to obtain a global health score and an abnormal diffusion risk score, and a Euclidean distance analysis is performed between embedded features based on the diffusion risk score to achieve highly sensitive prediction of abnormal diffusion chains.

[0104] Specifically, this step inherits the collaborative graph G(V,E,C) generated in step 3, and the input includes the equipment health status code z i (as node features), dynamic collaborative edge weight e i,j (from the “Collaborative Risk-Driven Mapping” in step 3), and the anomaly sensitivity factor from step 1 (Has participated in edge feature generation in step 2). This step designs the “Collaborative Health Prediction and Abnormal Diffusion Joint Evaluation Model” M in the cloud. GNN , specially tailored for the multi-equipment collaborative relationship and abnormal propagation characteristics of high-speed wire harness production lines.

[0105] Considering the unique characteristics of industry scenarios, high-speed wire harness production lines face challenges such as intensive process collaboration, fast cycle times, and the susceptibility of local anomalies to cascading failures. Existing health prediction solutions often focus solely on the health of individual devices or lack systematic quantification of the risk of anomaly diffusion. This step innovatively proposes a graph neural network mechanism combining "sensitivity modulation and collaborative propagation regularization." While maintaining the propagation capabilities of graph embedding features, it specifically strengthens the joint modeling of the "highly sensitive node → anomaly diffusion" feature, specifically targeting the industry's pain points.

[0106] M GNN The joint graph convolution mechanism of "node-sensitive fusion-cooperative risk modulation" is adopted. On the basis of standard graph neural networks, anomaly sensitivity and cooperative propagation regularization are introduced. This is specifically manifested in two aspects:

[0107] First, the node side modulates the information transmission weight of high-sensitivity nodes;

[0108] Second, the side uses the dynamic edge weight e i,j Further distinguish the amplifying effects of different synergistic relationships on abnormal diffusion.

[0109] In its implementation, the graph convolution operation is based on the following innovative “risk-sensitive modulation convolution unit”:

[0110]

[0111] in, represents the node embedding feature of device i at layer l, e i,j Indicates the collaborative edge weight calculated in step 3, dynamically reflecting C i and C j synergistic risks; represents the abnormal sensitivity of device i, from step 1; γ represents the abnormal sensitivity modulation factor, which is used to control the diffusion weight of highly sensitive devices; σ(·) represents the nonlinear activation function (such as ReLU); W (l) Represents the trainable parameter matrix of the lth layer.

[0112] Compared with conventional graph convolution, this formula innovatively transforms As a diffusion modulation item, the abnormal risk of highly sensitive equipment (such as a wire arrangement machine under high-speed conditions) is given a higher weight in the feature propagation process, which truly simulates the effect of "high-sensitive equipment abnormality → abnormal chain diffusion" in the industry scenario, and conforms to the actual characteristics of high coupling of multiple devices in high-speed wire harness production lines.

[0113] After completing the multi-layer graph convolution operation, the model outputs two core indicators at the node layer of the graph:

[0114] H i: The global health score of device i, which measures the health status of the device in the collaborative environment;

[0115] R i : The abnormal diffusion risk score of device i, reflecting the potential risk contribution of the abnormality of device i to its direct downstream nodes.

[0116] In order to achieve highly sensitive prediction of abnormal diffusion chains, an innovative "cooperative propagation regularization term" Ψ is proposed i,j , in calculating R i The “cross-layer collaborative transmission risk” mechanism is introduced, and the design is as follows:

[0117]

[0118] Among them, R i represents the abnormal diffusion risk of equipment i; H j represents the health score of neighbor device j; λ represents the coefficient of the cooperative propagation regularization term; It represents the Euclidean distance between the final embedded features of device i and neighbor j, reflecting the difference between the two in the feature space. The greater the difference, the higher the propagation potential.

[0119] Through this regularization mechanism, the system can automatically perceive the difference in health status between device i and downstream device j. When there is a transmission chain from "healthy device to vulnerable device", the system automatically amplifies R i , improve the detection capability of abnormal diffusion chain, and adapt to the industry pain point of "uneven equipment load distribution → easy spread of abnormalities" in the high-speed wiring harness industry.

[0120] Take the actual scenario as an example, the wiring machine is in a highly sensitive state, At the same time, its H 排线 =0.6, downstream welding equipment H 焊接 =0.4, and the difference in embedding features between the two is Ψ 排线,焊接 =0.2, then the R of the wire traversing machine 排线 The risk of abnormal spread will be amplified, prompting the subsequent scheduling system to pay special attention to this link.

[0121] S5. Analyze the global health score and anomaly spread risk score, and generate a grading strategy based on the device's position in the production line, the current production pace, and the device's optimization priority. Perform an anomaly analysis operation based on the dynamic collaborative graph. When a device is identified as a high-priority anomaly source, the system automatically selects associated downstream devices based on their link position in the dynamic collaborative graph to synchronously execute mitigation actions.

[0122] Specifically, this step receives the global health score H from step 4 i and abnormal diffusion risk R i, responsible for completing the "cloud dynamic scheduling strategy generation" and "edge control instruction issuance" based on these two core indicators, and realizing intelligent optimization control under the risk of equipment health status and coordinated abnormal diffusion. This step no longer involves any model building tasks and focuses on strategy generation and execution. In view of the characteristics of "multi-process high coordination + chain risk easy diffusion" of high-speed wire harness production lines, this step designs a "risk-driven hierarchical dynamic control mechanism" M OPT , in order to achieve flexible scheduling and linkage mitigation under different risk levels.

[0123] Cloud System M OPT H of device i i and R i As input, combined with the equipment's position in the production line (upstream / downstream) and the current production rhythm, three hierarchical strategies are generated:

[0124] High priority strategy: When H i Significantly below the threshold or R i If the risk exceeds the spread risk threshold, the system triggers the "local + coordinated mitigation" strategy;

[0125] Medium priority strategy: When H i Lower but R i If the condition is within the normal range, the system triggers the "single device self-healing" strategy;

[0126] Low priority strategy: When H i With R i All are within the controllable range. The system does not issue dynamic adjustments but only issues early warnings.

[0127] In order to achieve dynamic quantification, the system designs a “joint risk scoring function” U i , directly facing scheduling decisions, without participating in the graph modeling process:

[0128] U i =α·(1-H i )+β·R i (8)

[0129] Among them, U i represents the optimization priority score of device i; H i 、R i Denote the health score and diffusion risk, respectively, from step 4; α and β denote the modulation factors used to balance the “local abnormality risk” and “co-diffusion risk”.

[0130] Cloud based on U i Generate specific strategies based on the level of

[0131] When U i >θ H (such as θ H=0.7): Issue an "emergency mitigation" strategy to coordinate and optimize the device and its downstream equipment, taking measures such as joint load reduction, equipment speed reduction, and load balancing;

[0132] When θ M i ≤θ H (such as θ M =0.4): The "single-point self-healing" strategy is issued, and only local control optimization is performed on device i (such as dynamic adjustment of motor load and servo parameter optimization);

[0133] When U i ≤θ M : Only a monitoring report is generated, prompting "Device is healthy but requires further observation", and the control strategy is not triggered.

[0134] In order to ensure the systematic nature of flexible collaborative scheduling, a "link cascade mitigation mechanism" is designed. That is, when a device is identified as a high-priority abnormal source, the system will automatically select the associated downstream devices based on its link position in G(V, E, C) to perform mitigation actions simultaneously. For example, the cable arranging machine is identified as U 排线 =0.75, while its downstream welding machine R 焊接 =0.5, the system will simultaneously generate a linkage control strategy for the two devices to improve the collaborative resilience of the production line. After the edge receives the control strategy issued by the cloud, it performs a lightweight "strategy execution operation" with the following functions:

[0135] Parsing control instructions (such as "reduce speed by 10%" or "transfer load to a spare station");

[0136] Execute strategies in real time and feed back execution results to the cloud;

[0137] Ensure that strategies are implemented quickly when the production line runs at high speed (sub-second beats) to maintain the continuity and safety of the production line.

[0138] For example, the wiring machine H i =0.5, R i =0.55, cloud-based judgment U i =0.73, triggering the "coordinated mitigation" strategy. The system issues a "local load reduction" instruction to the cable arrangement machine and simultaneously issues a "early entry into inspection and maintenance" mitigation action to the downstream cable cutting machine to prevent the spread of high-risk link failures.

[0139] In one or more embodiments, Figure 2 As shown, a device dynamic management system based on real-time data collection and intelligent analysis is disclosed, the system comprising:

[0140] Production line data acquisition module 101, used to acquire real-time data of the high-speed wire harness assembly production line and perform pre-processing;​

[0141] The edge self-supervision module 102 is used to input the pre-processed data into the trained edge self-supervision model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data;

[0142] The cloud-based collaborative graph module 103 is configured to construct a dynamic collaborative graph for a high-speed wire harness production line on a cloud platform based on the health status code of the single device and the collaborative risk characteristics;

[0143] The GNN prediction module 104 is used to perform convolution operations based on the dynamic collaborative graph and the joint graph convolutional network to obtain a global health score and anomaly diffusion risk score, and perform Euclidean distance analysis between embedded features based on the diffusion risk score to highly sensitively predict anomaly diffusion chains;

[0144] The optimization execution module 105 is configured to analyze the global health score and anomaly diffusion risk score, generate a grading strategy based on the equipment's position in the production line, the current production pace, and the equipment's optimization priority, and perform anomaly analysis operations in conjunction with the dynamic collaboration graph. When a device is determined to be a high-priority anomaly source, the system automatically selects associated downstream devices based on their link position in the dynamic collaboration graph to synchronously execute mitigation actions.

[0145] The pre-processed data includes a weighted fusion feature vector of the current period of the device and a sensitivity factor of the current feature fluctuation of the device;

[0146] The trained edge self-supervision model includes at least an anomaly sensitivity modulation term and a collaborative regularization term. The anomaly sensitivity modulation term is used to control the degree of improvement of the anomaly separation capability of high-risk equipment; the collaborative regularization term is used to measure the degree of deviation between the characteristics of the equipment and the historical collaborative fault graph of the downstream equipment.

[0147] It is worth noting that the specific workflow of the equipment dynamic management system based on real-time data collection and intelligent analysis provided by an embodiment of the present invention is the same as the process of the equipment dynamic management method based on real-time data collection and intelligent analysis described in the above embodiment, and will not be repeated here.

[0148] Compared with the prior art, the embodiment of the present invention provides a dynamic equipment management based on real-time data acquisition and intelligent analysis, which obtains real-time data of a high-speed wire harness assembly production line and performs preprocessing; the preprocessed data is input into a trained edge self-supervisory model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data; a dynamic collaborative graph for a high-speed wire harness production line is constructed on a cloud platform based on the single device health status code and collaborative risk characteristics; based on the dynamic collaborative graph, a convolution operation is performed in combination with a joint graph convolutional network to obtain a global health score and anomaly diffusion risk score, and a Euclidean distance analysis is performed between embedded features based on the diffusion risk score to make a highly sensitive prediction of the anomaly diffusion chain; the global health score and anomaly diffusion risk score are analyzed, and a grading strategy is generated in combination with the position of the equipment in the production line, the current production rhythm, and the optimization priority of the equipment, and anomaly analysis operations are performed in combination with the dynamic collaborative graph; when a device is determined to be a high-priority anomaly source, the system automatically selects the associated downstream devices based on its link position in the dynamic collaborative graph to synchronously execute mitigation actions;

[0149] An embodiment of the present invention further provides a device for dynamic management of equipment based on real-time data collection and intelligent analysis, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the embodiment of the method for dynamic management of equipment based on real-time data collection and intelligent analysis are implemented, for example: Figure 1 or, the processor implements the functions of the modules in the above-mentioned system embodiments when executing the computer program.

[0150] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the device for dynamic device management based on real-time data collection and intelligent analysis.

[0151] The device for dynamic device management based on real-time data acquisition and intelligent analysis can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The device for dynamic device management based on real-time data acquisition and intelligent analysis can include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the device for dynamic device management based on real-time data acquisition and intelligent analysis can also include input / output devices, network access devices, buses, and the like.

[0152] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASAC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the device dynamic management system based on real-time data acquisition and intelligent analysis, and utilizes various interfaces and lines to connect the various parts of the device dynamic management system based on real-time data acquisition and intelligent analysis.

[0153] The memory can be used to store the computer program and / or module, and the processor realizes the various functions of the device dynamic management device based on real-time data acquisition and intelligent analysis by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the operation of the air-conditioning controller, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMedaaCard, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0154] Wherein, if the module for integrating dynamic device management based on real-time data acquisition and intelligent analysis is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0155] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0156] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A device dynamic management method based on real-time data collection and intelligent analysis, characterized in that: The method comprises: Acquire real-time data from high-speed wire harness assembly lines and perform pre-processing; Inputting the preprocessed data into a trained edge self-supervisory model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data; Based on the single-device health status coding and collaborative risk characteristics, a dynamic collaborative graph for a high-speed wire harness production line is constructed on a cloud platform; Based on the dynamic collaborative graph, combined with the joint graph convolutional network, a convolution operation is performed to obtain a global health score and anomaly diffusion risk score. The Euclidean distance analysis between embedded features is performed based on the diffusion risk score to achieve highly sensitive prediction of anomaly diffusion chains. The global health score and anomaly spread risk score are analyzed, and a grading strategy is generated based on the device's position in the production line, the current production rhythm, and the device's optimization priority. The system then performs anomaly analysis operations in conjunction with the dynamic collaborative graph. When a device is identified as a high-priority anomaly source, the system automatically selects associated downstream devices based on their link position in the dynamic collaborative graph to synchronously execute mitigation actions. The pre-processed data includes a weighted fusion feature vector of the current period of the device and a sensitivity factor of the current feature fluctuation of the device; The trained edge self-supervision model includes at least an anomaly sensitivity modulation term and a collaborative regularization term. The anomaly sensitivity modulation term is used to control the degree of improvement of the anomaly separation capability of high-risk equipment; the collaborative regularization term is used to measure the degree of deviation between the characteristics of the equipment and the historical collaborative fault graph of the downstream equipment.

2. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 1, characterized in that: The real-time data includes the current equipment's production rhythm, load status and historical abnormality rate, as well as the equipment's vibration, temperature and pressure signals; Wherein, the preprocessing includes: Dynamically adjust the time window lengths of different signals according to the production rhythm, load status, and historical abnormality rate of the current equipment; Obtaining a weighted fusion feature vector of the current period of the device by weighted fusion according to the vibration, temperature, and pressure signals of the device; A dynamic weighted standard analysis is performed based on the weighted fusion feature vector of the current period of the device to obtain a sensitivity factor of the current feature fluctuation of the device for abnormality warning.

3. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 1, characterized in that: The trained edge self-supervised model adopts an improved contrastive learning framework and is deployed in edge devices.

4. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 3 is characterized in that: The trained edge self-supervised model, loss function Expressed as: in, Training target for edge self-supervised model; z i Encode the health status of device i, which represents the potential health distribution of the device; For device i, feature F i A weakly enhanced version of is the abnormal sensitivity factor of device i; β is the sensitivity modulation factor, which controls the degree of improvement of the abnormal separation capability of high-risk devices; is the collaborative regularization term, which measures the characteristics of device i and the historical collaborative failure graph G of downstream devices hist The degree of deviation; η is the collaborative regularization weight; The collaborative regularization term is related to the distance between the health status code of the device and the historical collaborative anomaly center.

5. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 1, characterized in that: The structure of the dynamic collaboration graph is: Node set V = {v i }: Each v i Represents production line equipment i, and the node feature is the health status code z i ; Edge set E={e i,j }: The synergistic edge between equipment i and equipment j. The strength of the edge is affected by the production process dependency and the synergistic risk characteristic C of equipment i. i , Equipment j Collaborative Risk Characteristics C j Dynamic impact; Collaborative risk set C = {C i }: Synchronously obtain from the edge side; Among them, the collaborative edge weight e i,j The dynamic generation of is completed by the following formula: Among them, e i,j is the edge weight between device i and device j; κ i,j is the static process dependency value in the production process; δ is the adjustment coefficient, which is set for the global production rhythm; C i 、C j The collaborative risk signature generated for the edge indicates the potential for device anomaly propagation.

6. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 5, characterized in that: The convolution operation of the joint graph convolutional network is specifically: Use node features as health status encoding z i as the node embedding feature of device i at layer 0; The sensitivity factor of the current characteristic fluctuation of the device is used as the diffusion modulation item; For the node embedding features of device i at layer 0, that is, node i, then aggregate the features of its neighbor node j, weight the neighbor features with the weight matrix, and introduce a diffusion modulation term to enhance or suppress the influence of neighbor information. Finally, the updated node features are output through the activation function σ; the updated node features include the global health score of the device and the abnormal diffusion risk score of the device.

7. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 6, characterized in that: The joint graph convolutional network also includes a collaborative propagation regularization term, which is used to output the abnormal diffusion risk score of the device by analyzing the Euclidean distance between the final embedded features of device i and its neighbor j; wherein the Euclidean distance reflects the difference between the two in the feature space, and the largest difference indicates the highest propagation potential.

8. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 1, characterized in that: The global health score and the abnormal diffusion risk score are analyzed, and three grading strategies are generated in combination with the position of the equipment in the production line and the current production rhythm: a high priority strategy, a medium priority strategy, and a low priority strategy.

9. The method for dynamic equipment management based on real-time data collection and intelligent analysis according to claim 8, characterized in that: After obtaining the three classification strategies, the following steps are also included: Obtain the global health score and abnormal spread risk score, and perform multi-factor weighted analysis to obtain the optimization priority score of the equipment; In combination with the optimization priority score of the device, a specific control strategy is generated; the specific control strategy includes: When the optimization priority score is greater than the highest threshold, an emergency mitigation strategy is issued to coordinate optimization of the device and its downstream devices, and joint measures are taken. When the optimization priority score is less than the highest threshold but greater than the middle threshold, a single-point self-healing policy is issued to perform local control optimization only on device i. When the optimization priority score is lower than the intermediate threshold, only a monitoring report is generated, indicating that the device is healthy but requires further observation, and the control policy is not triggered. After receiving the control strategy, the edge performs lightweight strategy execution operations, including: parsing control instructions; executing the strategy in real time and feeding back the execution results to the cloud; ensuring that the strategy is quickly implemented under the high-speed operation of the production line, maintaining the continuity and safety of the production line.

10. A dynamic equipment management system based on real-time data collection and intelligent analysis, characterized in that: The system comprises: Production line data acquisition module, used to obtain real-time data of high-speed wire harness assembly production line and perform pre-processing; An edge self-supervision module is used to input the pre-processed data into a trained edge self-supervision model to generate a single device health status code and collaborative risk characteristics corresponding to the real-time data; A cloud-based collaborative graph module is used to construct a dynamic collaborative graph for a high-speed wire harness production line on a cloud platform based on the health status code of the single device and the collaborative risk characteristics; A GNN prediction module is used to perform convolution operations based on the dynamic collaborative graph and a joint graph convolutional network to obtain a global health score and anomaly diffusion risk score, and perform Euclidean distance analysis between embedded features based on the diffusion risk score to achieve highly sensitive prediction of anomaly diffusion chains; An optimization execution module is used to analyze the global health score and anomaly diffusion risk score, and generate a classification strategy based on the device's position in the production line, the current production rhythm, and the device's optimization priority, and perform anomaly analysis operations in conjunction with the dynamic collaboration graph. When a device is determined to be a high-priority anomaly source, the system automatically selects associated downstream devices based on their link position in the dynamic collaboration graph to synchronously execute mitigation actions. The pre-processed data includes a weighted fusion feature vector of the current period of the device and a sensitivity factor of the current feature fluctuation of the device; The trained edge self-supervision model includes at least an anomaly sensitivity modulation term and a collaborative regularization term. The anomaly sensitivity modulation term is used to control the degree of improvement in the anomaly separation capability of high-risk equipment; the collaborative regularization term measures the degree of deviation between the characteristics of the equipment and the historical collaborative fault graph of the downstream equipment.

Images