Precise part production equipment fault monitoring method and system based on digital twinning

Abstract

The invention relates to the related technical field of intelligent production monitoring, in particular to a precision part production equipment fault monitoring method and system based on digital twinning, and the method comprises the steps: collecting equipment state data, constructing an equipment digital twinning model, and carrying out the multi-dimensional simulation deduction through combining with the equipment state data, determining a key component early warning threshold value and a fault response strategy; deploying edge monitoring nodes, setting degradation sensitive features, and generating a fault evolution prediction map; meanwhile, a fault conduction network is constructed, and fault influence range marking reminding is carried out. The technical problems that monitoring depends on single sensing data and a fixed threshold value, internal correlation between state abnormity and process fluctuation cannot be recognized, and the production stability of precision parts is insufficient are solved, tool abrasion loss signals are collected, abnormal intervals are matched through cosine similarity, a fault evolution prediction map is generated, and the fault evolution prediction accuracy is improved. The accuracy and collaboration of fault monitoring are improved, and reliable guarantee is provided for the production stability of precise parts.

Description

Technical Field

[0001] This invention relates to the field of intelligent production monitoring technology, specifically to a method and system for fault monitoring of precision component manufacturing equipment based on digital twins. Background Technology

[0002] The structures of precision machining equipment such as five-axis machining centers and high-speed milling machines are becoming increasingly complex. The operating status of key components such as spindles, cutting tools, and guideways is affected by multiple factors such as machining load, material properties, and process parameters, making them prone to latent degradation during long-term operation. However, current fault monitoring of precision component production equipment relies heavily on single sensor data, which cannot identify the intrinsic relationship between abnormal status and process fluctuations, and is prone to missing early latent fault signals. In addition, the equipment operating conditions and production cycle lack dynamic adaptation and cannot be autonomously iterated and optimized through actual operation and maintenance data, resulting in a high rate of false alarms and missed faults, which restricts the stability of precision component production.

[0003] In summary, existing technologies suffer from the following technical problems: they rely on single sensor data and fixed thresholds for monitoring, which fails to identify the intrinsic relationship between abnormal conditions and process fluctuations, resulting in insufficient production stability of precision components. Summary of the Invention

[0004] This application provides a method and system for fault monitoring of precision component manufacturing equipment based on digital twins, aiming to solve the technical problems of existing technologies that rely on single sensor data and fixed thresholds for monitoring, which cannot identify the intrinsic relationship between abnormal conditions and process fluctuations, resulting in insufficient production stability of precision components.

[0005] In view of the above problems, the technical solution to achieve the present application is as follows: In a first aspect, this application provides a method for fault monitoring of precision component manufacturing equipment based on digital twins. The method includes: collecting equipment status data, including spindle vibration amplitude and tool temperature field, under the same time stamp based on sensor monitoring data and process execution information; constructing a digital twin model of the equipment based on the sensor monitoring data and a fault mode knowledge base, and performing multi-dimensional simulation and deduction in conjunction with the equipment status data to determine the early warning threshold and fault response strategy for key components; deploying edge monitoring nodes to collect tool wear signals, combining the machining accuracy decay trend within the production cycle of the process execution information, matching abnormal wear intervals through cosine similarity, setting degradation-sensitive features under an attention mechanism, generating a fault evolution prediction map within a time sliding window, and performing adaptive early warning level matching and control command compensation; simultaneously, constructing a fault propagation network based on the early warning threshold and fault response strategy for key components, and marking and reminding users of the fault impact range.

[0006] Preferably, based on the fault propagation network, key fault indicators including bearing temperature rise rate and guide rail clearance change rate are extracted; based on the key fault indicators, combined with equipment operating conditions and process accuracy requirements, dynamic scheduling parameters and priority response rules for maintenance resources are configured.

[0007] Preferably, the maintenance resources interact with the digital twin model of the equipment in real time; the scheduling path of the maintenance resources is dynamically adapted to the equipment downtime window, and the fault mode knowledge base is iteratively updated using a deep reinforcement learning algorithm.

[0008] Preferably, a dual-dimensional reward function is set, wherein the first dimension is the matching degree between the fault prediction deviation rate and the actual fault degree, and the second dimension is the maintenance response time. Based on the dual-dimensional reward function, a deep Q-network is used to learn the fault identification strategy in the discrete fault type space.

[0009] Preferably, the spindle vibration amplitude and tool temperature field are bound to the production process nodes and processing batch information to establish a state-process correlation matrix; based on the state-process correlation matrix, the state differences of the same production equipment at different processing stages are identified, and benchmark state feature values ​​are extracted; the benchmark state feature values ​​are used to automatically identify abnormal processes and push them to the equipment health management terminal.

[0010] Preferably, a material fatigue curve library for precision components is embedded in the digital twin model of the equipment. The material fatigue curve library contains life decay parameters under different stress amplitudes. Based on the material fatigue curve library, the stress accumulation process of key components under different processing loads and different cutting path combinations is simulated to generate a life consumption Pareto front and a machining accuracy Pareto front. According to the production accuracy priority, the optimal equilibrium point in the non-dominated solution set is selected from the life consumption Pareto front and the machining accuracy Pareto front.

[0011] Preferably, spindle speed, feed rate, and tool life allowance are used as variable parameters to set constraints in the digital twin model of the equipment; based on the constraints, a valid sample set is determined, and non-dominated sorting is performed to obtain the non-dominated solution set.

[0012] Preferably, a wavelet transform unit is integrated on the edge monitoring node to obtain a subset of degradation-sensitive features; the subset of degradation-sensitive features is fused with the cutting parameters in the process execution information and used as the input of the gated loop unit network to predict the wear change trend within the production cycle of the process execution information, and to determine whether to trigger a tool change command based on the normal wear range.

[0013] Preferably, the tool wear signal is decomposed into feature components of multiple frequency bands; energy entropy and kurtosis features are extracted from the feature components of the multiple frequency bands to generate a time-frequency domain sensitive feature vector, and a subset of degradation sensitive features strongly correlated with tool degradation is selected.

[0014] In a second aspect, this application provides a fault monitoring system for precision component manufacturing equipment based on digital twins. The system includes: a data collection module: collecting equipment status data, including spindle vibration amplitude and tool temperature field, under the same time stamp based on sensor monitoring data and process execution information; a simulation and deduction module: constructing a digital twin model of the equipment based on the sensor monitoring data and a fault mode knowledge base, and performing multi-dimensional simulation and deduction based on the equipment status data to determine the early warning threshold and fault response strategy for key components; a fault evolution module: deploying edge monitoring nodes to collect tool wear signals, combining the machining accuracy decay trend within the production cycle of the process execution information, matching abnormal wear intervals through cosine similarity, setting degradation-sensitive features under an attention mechanism, generating a fault evolution prediction map within a time sliding window, and performing adaptive early warning level matching and control command compensation; and a fault marking and reminder module: simultaneously constructing a fault propagation network based on the early warning threshold and fault response strategy for key components, and marking and reminding users of the fault impact range.

[0015] In summary, one or more technical solutions provided in this application achieve the technical effect of collecting tool wear signals, matching abnormal intervals through cosine similarity, generating fault evolution prediction maps, improving the accuracy and synergy of fault monitoring, and providing reliable assurance for the production stability of precision components. Attached Figure Description

[0016] Figure 1 This application provides a flowchart illustrating a method for fault monitoring in precision component manufacturing equipment based on digital twins.

[0017] Figure 2 This application provides a structural schematic diagram of a fault monitoring system for precision component manufacturing equipment based on digital twins.

[0018] Explanation of reference numerals in the attached diagram: Data collection module M100, simulation and deduction module M200, fault evolution module M300, fault marking and reminder module M400. Detailed Implementation

[0019] Example 1: The present application will be described in detail below with reference to the accompanying drawings, as follows... Figure 1 As shown, this application provides a fault monitoring method for precision component manufacturing equipment based on digital twins, wherein the method includes: S1: Based on sensor monitoring data and process execution information, collect equipment status data including spindle vibration amplitude and tool temperature field under the same timestamp; S2: Based on the sensor monitoring data and fault mode knowledge base, construct a digital twin model of the equipment, and combine the equipment status data to perform multi-dimensional simulation and deduction to determine the early warning threshold of key components and fault response strategies.

[0020] Specifically, sensor monitoring data refers to real-time data collected by various sensors such as vibration sensors and temperature sensors during equipment operation, reflecting the physical state of the equipment; process execution information refers to the specific process parameters executed by the equipment during production, including processing speed, feed rate, and depth of cut, which are closely related to the equipment status; the same timestamp constraint is used to ensure that the collected equipment status data and process execution information are synchronized in time; the equipment digital twin model refers to the ability to reflect the real-time operating status of physical equipment by integrating sensor monitoring data and a fault mode knowledge base, and to be used to simulate and predict equipment behavior; multi-dimensional simulation and inference refers to the comprehensive evaluation of equipment status by simulating different working conditions, loads, and processing paths in the digital twin model, in order to determine the early warning thresholds and fault response strategies for key components.

[0021] Execution steps: Synchronously collect sensor monitoring data and process execution information from the equipment to ensure data consistency over time. Furthermore, deeply bind the physical state of the equipment, including spindle vibration amplitude and tool temperature field, with process parameters such as machining speed and feed rate during production. In addition, in precision machining equipment, spindle vibration amplitude may increase significantly when machining high-hardness materials. By analyzing the correlation with process execution information, including cutting depth and feed rate, it is possible to accurately identify whether this state change is a normal machining fluctuation or a potential fault signal.

[0022] By utilizing this synchronized data to construct a digital twin model of the equipment and conducting multi-dimensional simulations, the digital twin model can simulate the equipment's operating status under different working conditions. Through integration with a fault mode knowledge base, it predicts potential fault modes. Furthermore, by simulating the cumulative effect of spindle vibration under different processing loads, it determines the types of faults that may occur when the spindle vibration amplitude exceeds a certain threshold. Based on these simulation results, early warning thresholds for key components are dynamically determined; for example, the early warning threshold for spindle vibration amplitude can be set to 1.5 times the normal processing fluctuation range. Simultaneously, corresponding fault response strategies are generated, such as automatically adjusting processing parameters or issuing maintenance reminders when the spindle vibration amplitude exceeds the early warning threshold. Preferably, the dynamic early warning and response strategy generation based on the digital twin model, compared to traditional fixed threshold methods, can more accurately predict faults, reduce false alarms and missed alarms, significantly improve the accuracy and coordination of fault monitoring in precision component production equipment, and provide strong support for production stability.

[0023] S3: Deploy edge monitoring nodes to collect tool wear signals, combine the machining accuracy decay trend within the production cycle of the process execution information, match the abnormal wear interval through cosine similarity, set the degradation sensitive features under the attention mechanism, generate a fault evolution prediction map within the time sliding window, and perform adaptive early warning level matching and control command compensation; S4: At the same time, construct a fault transmission network based on the early warning threshold and fault response strategy of the key components, and mark and remind the fault impact range.

[0024] Specifically, edge monitoring nodes refer to monitoring units deployed at the edge of equipment, capable of collecting local state data of the equipment in real time, such as tool wear signals, and featuring low latency and high real-time performance; cosine similarity matching is used to measure the similarity between two vectors, comparing the similarity between tool wear signals and normal wear patterns to identify abnormal wear intervals; attention mechanisms refer to improving the processing efficiency of important information by automatically identifying and focusing on key features in the data; time sliding windows refer to extracting local features by sliding a window on time series data, used to analyze fault evolution trends; and fault propagation networks are models used to analyze and predict the propagation path and impact range of faults within the equipment or system.

[0025] Execution steps: Monitoring nodes are deployed at the equipment edge to collect tool wear signals in real time. Combined with the machining accuracy decay trend within the production cycle of the process execution information, abnormal wear intervals are matched using cosine similarity. Furthermore, the feature vector of the normal tool wear pattern within a certain production cycle is denoted as... The feature vector of the real-time acquired tool wear signal is denoted as By calculating cosine similarity ,when When the tool wear falls below a certain set threshold, it can be determined that the tool wear is abnormal. Specifically, the set threshold can be set to 0.8. Further, in the tool wear signal, the attention mechanism is used to identify features that are strongly correlated with tool wear, such as an increase in high-frequency vibration components or an abnormal increase in temperature. After these features are extracted, they are used to generate a fault evolution prediction map within a time sliding window. Specifically, a sliding window with a length of 10 minutes is set on the time series data, and the sliding window is moved every minute to analyze the changing trend of tool wear characteristics within each window and predict possible future faults.

[0026] Based on critical component early warning thresholds and fault response strategies, a fault propagation network is constructed. Furthermore, when the spindle vibration amplitude exceeds the early warning threshold, the fault propagation network analyzes other components that may be affected by the fault, including guide rails and cutting tools. The scope of the fault's impact is marked as an alert, and in the equipment's digital twin model, affected components are marked in red to remind operators to take timely measures. Through these steps, real-time monitoring and fault prediction of tool wear are achieved. Simultaneously, the fault propagation network predicts and alerts to the cascading effects of faults. This method, based on edge computing and attention mechanisms, compared to traditional centralized monitoring, can more quickly capture early fault signals, reduce false alarms and missed alarms, and significantly improve the accuracy and synergy of fault monitoring in precision component manufacturing equipment, providing a reliable guarantee for production stability.

[0027] Furthermore, based on the aforementioned critical component warning thresholds and fault response strategies, a fault propagation network is constructed to mark and alert on the fault impact range. The method of this application also includes: Based on the fault propagation network, key fault indicators, including bearing temperature rise rate and guide rail clearance change rate, are extracted; based on the key fault indicators, combined with equipment operating conditions and process accuracy requirements, dynamic scheduling parameters and priority response rules for maintenance resources are configured.

[0028] Specifically, fault propagation networks are models used to analyze the propagation path and impact range of faults within equipment or systems, and can predict the impact of faults on other components; critical fault indicators refer to parameters that have a significant impact on the operating status of equipment during fault propagation, including bearing temperature rise rate and guide rail clearance change rate, used to reflect the severity and development trend of faults; dynamic scheduling parameters refer to maintenance resource allocation parameters that are adjusted in real time according to equipment operating status and fault indicators, used to optimize the efficiency of maintenance resource utilization; priority response rules refer to determining the priority order of maintenance responses based on the severity of faults and their impact on production, ensuring that critical faults can be handled in a timely manner.

[0029] Execution steps: Based on the fault propagation network, key fault indicators are extracted, including bearing temperature rise rate and guide rail clearance change rate. Furthermore, in precision machining equipment, when a spindle malfunctions, fault propagation network analysis reveals an abnormal increase in bearing temperature rise rate and fluctuations in guide rail clearance change rate. These key fault indicators reflect the fault propagation path and impact range, providing a basis for subsequent maintenance resource scheduling. Based on these key fault indicators, combined with equipment operating conditions and process precision requirements, dynamic scheduling parameters and priority response rules for maintenance resources are configured. Furthermore, the equipment operating conditions require high-precision machining; when the bearing temperature rise rate exceeds a certain threshold, it leads to a decrease in machining accuracy. For example, if the threshold for the bearing temperature rise rate is set to 0.5℃ per minute, the allocation of maintenance resources is automatically adjusted according to the pre-set dynamic scheduling parameters, such as increasing the power of the cooling system. Simultaneously, according to the priority response rules, faults with abnormal bearing temperature rise rates are marked as high priority to prevent further impact on production.

[0030] Through the above steps, rapid response to faults and precise maintenance resource scheduling are achieved. Furthermore, by optimizing dynamic scheduling parameters, abnormal bearing temperature rise rates can be effectively controlled in a short time, shortening the response time. At the same time, by using priority response rules, the processing time for critical faults is reduced, significantly improving the operational stability and production efficiency of the equipment. The dynamic scheduling method based on fault propagation networks and key fault indicators can effectively reduce the impact of faults on production and improve the maintenance efficiency and production stability of precision component production equipment.

[0031] Furthermore, the method of this application also includes: The maintenance resources interact with the digital twin model of the equipment in real time; the scheduling path of the maintenance resources is dynamically adapted to the equipment downtime window, and the fault mode knowledge base is iteratively updated using a deep reinforcement learning algorithm.

[0032] Specifically, real-time interaction between maintenance resources and equipment digital twin models refers to the real-time information exchange and feedback between maintenance resources such as maintenance tools and spare parts and the digital twin models of equipment; the digital twin model can dynamically adjust maintenance strategies based on the status and location of maintenance resources; maintenance resources can also prepare in advance based on the predictive information provided by the digital twin model; dynamic adaptation of scheduling paths and equipment downtime windows refers to dynamically adjusting the scheduling paths of maintenance resources based on the actual operating status of the equipment and the planned downtime, ensuring that maintenance work can be completed efficiently within the equipment downtime window and reducing equipment downtime; deep reinforcement learning algorithms refer to the use of intelligent agents to learn through trial and error in the environment, optimizing their behavioral strategies based on reward functions, and iteratively updating the fault mode knowledge base to improve the accuracy of fault prediction and the timeliness of response; the fault mode knowledge base is a database that stores equipment fault modes, features, and response strategies to support fault monitoring and prediction.

[0033] Execution steps: Real-time interaction between maintenance resources and the digital twin model of the equipment. Furthermore, when the digital twin model predicts an impending failure of the equipment spindle, it immediately notifies the maintenance resource management system. Based on the failure prediction information provided by the digital twin model, such as the probability of failure and the estimated time, the maintenance resource management system dynamically adjusts the allocation of maintenance resources and prepares spare parts for the equipment spindle in advance to ensure preventative maintenance can be performed before a failure occurs. The scheduling path of maintenance resources is dynamically adapted to the equipment downtime window. For example, if the equipment is scheduled to be shut down for maintenance at midnight, the maintenance resource management system will ensure that it can arrive at the site before the downtime and complete the maintenance work within the downtime, based on the downtime window. In this way, equipment downtime can be minimized and equipment utilization can be improved.

[0034] A deep reinforcement learning algorithm is used to iteratively update the fault mode knowledge base, with a two-dimensional reward function. The first dimension is the matching degree between the fault prediction deviation rate and the actual fault severity, and the second dimension is the maintenance response timeliness. Based on these two dimensions, a deep Q-network learns fault identification strategies in a discrete fault type space. Through continuous learning and optimization, the fault mode knowledge base can be autonomously updated based on actual operation and maintenance data, improving the accuracy of fault prediction and the timeliness of response. Through these steps, efficient management of maintenance resources and continuous optimization of fault prediction are achieved. Specifically, real-time interaction between maintenance resources and the digital twin model reduces equipment downtime; iterative updates to the fault mode knowledge base using deep reinforcement learning improve fault prediction accuracy; and maintenance strategies based on real-time interaction and intelligent learning effectively improve equipment operational stability and production efficiency, providing strong support for the efficient operation of precision component manufacturing equipment.

[0035] Furthermore, the method of this application includes iteratively updating the fault mode knowledge base using a deep reinforcement learning algorithm: A two-dimensional reward function is set up, where the first dimension is the matching degree between the fault prediction deviation rate and the actual fault degree, and the second dimension is the maintenance response time. Based on the two-dimensional reward function, a deep Q-network is used to learn the fault identification strategy in the discrete fault type space.

[0036] Specifically, the two-dimensional reward function includes a reward mechanism with two main evaluation metrics to guide the learning process of the deep Q-network. The first dimension is the matching degree between the fault prediction bias rate and the actual fault severity, measuring the accuracy of fault prediction; the second dimension is the maintenance response timeliness, measuring the timeliness of the maintenance response. The fault prediction bias rate refers to the difference between the predicted probability or time of a fault occurrence and the actual probability or time of a fault occurrence. A lower bias rate indicates more accurate prediction. The maintenance response timeliness refers to the time interval from the occurrence of a fault to the start of a maintenance response; a shorter response timeliness indicates a more timely maintenance response. The deep Q-network is used to learn optimal decision-making strategies in complex environments, specifically, learning fault identification strategies in a discrete fault type space. The discrete fault type space contains the state space of all possible fault types, with each fault type corresponding to a discrete state.

[0037] Execution Steps: A two-dimensional reward function is set to guide the learning process of the deep Q-network. The first dimension is the match between the fault prediction bias rate and the actual fault severity, used to measure the accuracy of fault prediction. For example, assuming the fault prediction bias rate is defined as the ratio of the absolute difference between the predicted fault occurrence time and the actual fault occurrence time to the actual fault occurrence time, when the bias rate is below a certain threshold, the first dimension of the reward function assigns a higher reward value, indicating accurate prediction. For example, the threshold for the bias rate can be set to 10%. The second dimension is maintenance response timeliness, used to measure the timeliness of maintenance response. For example, maintenance response timeliness is defined as the time interval from the fault occurrence to the start of the maintenance response. When the response time is less than a certain threshold, the second dimension of the reward function assigns a higher reward value, indicating timely response. For example, the threshold for response time can be set to 10 minutes.

[0038] Based on a two-dimensional reward function, a deep Q-network is used to learn a fault identification strategy in a discrete fault type space. Specifically, the discrete fault type space includes several common fault types such as spindle faults, tool wear, and guideway faults. Through interaction with the environment, the deep Q-network learns to take optimal maintenance instructions or adjust machining parameters under different fault types and equipment operating conditions. During the learning process, the network continuously adjusts its strategy based on feedback from the two-dimensional reward function to improve the accuracy of fault prediction and the timeliness of maintenance response. Through these steps, the fault identification strategy based on deep reinforcement learning can significantly improve the efficiency of equipment fault monitoring and maintenance, reduce downtime, improve production stability, and achieve intelligent optimization of fault prediction and maintenance response.

[0039] Furthermore, the method of this application for collecting equipment status data, including spindle vibration amplitude and tool temperature field, under the same timestamp, includes: The spindle vibration amplitude and tool temperature field are bound to the production process nodes and processing batch information to establish a state-process correlation matrix. Based on the state-process correlation matrix, the state differences of the same production equipment at different processing stages are identified, and the baseline state feature values ​​are extracted. The baseline state feature values ​​are used to automatically identify abnormal processes and push them to the equipment health management terminal.

[0040] Specifically, the state-process correlation matrix is ​​used to represent the relationship between equipment state data, including spindle vibration amplitude and tool temperature field, and production process nodes and processing batch information. Each row represents a process node or processing batch, and each column represents an equipment state parameter. The baseline state characteristic value refers to the state characteristic value of the equipment during normal operation at different processing stages, which is used as a reference standard to identify abnormal states. Automatic identification of abnormal state processes refers to automatically identifying process nodes with abnormal equipment states by comparing the real-time collected equipment state data with the baseline state characteristic value. The equipment health management terminal is a terminal device used to monitor and manage the health status of equipment, and can receive and display equipment status information and abnormal alarms.

[0041] Execution steps: Bind equipment status data such as spindle vibration amplitude and tool temperature field with production process nodes and processing batch information to establish a status-process correlation matrix. For example, a precision machining equipment has multiple main process nodes, each corresponding to different processing parameters and equipment status. By collecting spindle vibration amplitude and tool temperature field data at each process node and binding these data with the corresponding process node and processing batch information, a status-process correlation matrix is ​​formed. The status-process correlation matrix can intuitively display the status changes of the equipment at different process nodes.

[0042] Based on the state-process correlation matrix, the state differences of the same production equipment at different processing stages are identified, and baseline state feature values ​​are extracted. By analyzing the state-process correlation matrix, it is found that there are significant differences in the spindle vibration amplitude and tool temperature field at different process nodes. Through statistical analysis, baseline state feature values ​​for each process node are extracted, such as the average and standard deviation of the spindle vibration amplitude and the average and standard deviation of the tool temperature field. These baseline state feature values ​​serve as reference standards for normal operation.

[0043] The system automatically identifies abnormal processes using baseline state characteristic values ​​and pushes the results to the equipment health management terminal. For example, if real-time equipment status data shows that the spindle vibration amplitude exceeds the threshold range of the baseline state characteristic values ​​(e.g., exceeding three times the standard deviation of the average), the system automatically identifies an abnormal state at that process node and pushes an alarm to the equipment health management terminal. The terminal receives and displays the alarm, reminding operators to check and handle it promptly. In these steps, by analyzing the state-process correlation matrix and baseline state characteristic values, the system identifies process nodes with abnormal equipment conditions. This data-binding and feature extraction-based method significantly improves the efficiency and accuracy of equipment status monitoring, reduces production interruptions caused by equipment failures, and enables real-time monitoring and anomaly identification of equipment status.

[0044] Furthermore, this application's method involves constructing a digital twin model of the equipment and performing multi-dimensional simulations based on the equipment status data to determine early warning thresholds for key components and fault response strategies. A material fatigue curve library for precision components is embedded in the digital twin model of the equipment. The material fatigue curve library contains life decay parameters under different stress amplitudes. Based on the material fatigue curve library, the stress accumulation process of key components under different processing loads and different cutting path combinations is simulated to generate the life consumption Pareto front and the processing accuracy Pareto front. According to the production accuracy priority, the optimal equilibrium point in the non-dominated solution set is selected from the life consumption Pareto front and the processing accuracy Pareto front.

[0045] Specifically, the material fatigue curve library stores fatigue life decay parameters for different materials under different stress amplitudes, describing the fatigue characteristics of materials under repeated loading; the Pareto front for life consumption refers to the set of solutions in multi-objective optimization where further optimization between life consumption and machining accuracy is not possible, and these solutions achieve the best balance between life consumption and machining accuracy; the Pareto front for machining accuracy also refers to the set of solutions where further optimization between machining accuracy and life consumption is not possible, and these solutions achieve the best balance between machining accuracy and life consumption; the non-dominated solution set refers to a set of solutions in multi-objective optimization where no single solution is superior to another in all objectives, and these solutions represent the best balance point between different objectives; the optimal balance point refers to the most suitable solution selected in the non-dominated solution set according to specific priorities or criteria, usually the best compromise under given conditions.

[0046] Execution steps: Embed a material fatigue curve library for precision components into the digital twin model of the equipment. For example, the key component of the equipment is the spindle, which is made of high-strength steel. The material fatigue curve library stores the fatigue life decay parameters of this material under different stress amplitudes, describing the fatigue characteristics of the material under repeated loading. Specifically, when the stress amplitude is 100MPa, the fatigue life of the material is 10^6 cycles; when the stress amplitude is 200MPa, the fatigue life is 10^5 cycles. Based on the material fatigue curve library, simulate the stress accumulation process of the key component under different processing loads and different cutting path combinations. Furthermore, the equipment has different processing loads and different cutting paths during the processing. Through the digital twin model, simulate the stress accumulation process of the spindle under these different working conditions, and calculate the stress accumulation value and life consumption under each working condition.

[0047] The Pareto front for lifetime consumption and the Pareto front for machining accuracy are generated. Using data obtained from simulation, the Pareto front between lifetime consumption and machining accuracy is plotted. Furthermore, under light load and straight cutting path, lifetime consumption is low but machining accuracy is high; while under heavy load and curved cutting path, lifetime consumption is high but machining accuracy is low. Based on these data, the Pareto front curve is generated, representing the solution set that cannot be further optimized between lifetime consumption and machining accuracy.

[0048] Based on the production accuracy priority, the optimal balance point is selected from the non-dominated solution set between the Pareto front for lifespan consumption and the Pareto front for machining accuracy. Furthermore, if the production accuracy priority is high, a solution with high machining accuracy and relatively low lifespan consumption is selected from the Pareto front as the optimal balance point. Specifically, there is a solution on the Pareto front with a machining accuracy of 95% and a lifespan consumption of 10%. This solution meets the production accuracy requirements while maintaining an acceptable lifespan consumption, and is therefore selected as the optimal balance point. Through these steps, by embedding a material fatigue curve library and simulating the stress accumulation process under different working conditions, the optimal balance point between machining accuracy and lifespan consumption is found. This method, based on a digital twin model and multi-objective optimization, can significantly improve equipment operating efficiency and production stability, reduce failures caused by component fatigue, increase production efficiency, and achieve optimized management of the lifespan and machining accuracy of key equipment components.

[0049] Furthermore, the method of this application includes: Spindle speed, feed rate, and tool life allowance are used as variable parameters, and constraints are set in the digital twin model of the equipment. Based on the constraints, a valid sample set is determined, and non-dominated solution set is obtained by non-dominated sorting.

[0050] Specifically, variable parameters refer to parameters that can be adjusted during the optimization process, including spindle speed, feed rate, and tool life margin. These parameters directly affect the machining process and equipment status. Constraints refer to the conditions that limit the range of variable parameters in the optimization problem, ensuring that the solution is feasible in practical applications. Specifically, the spindle speed cannot exceed its rated speed, and the tool life margin cannot be negative. The effective sample set refers to the set of samples that satisfy all constraints, used for further optimization analysis. Non-dominated sorting is used for multi-objective optimization, which means dividing the sample set into different levels, where samples in each level are not better than samples in other levels for any objective. The non-dominated solution set refers to the optimal solution set obtained after non-dominated sorting, where these solutions achieve the best balance among multiple objectives.

[0051] Execution steps: Spindle speed, feed rate, and tool life allowance are used as variable parameters, which directly affect machining efficiency and equipment life. Constraints are set in the equipment's digital twin model to ensure that the values ​​of these parameters are feasible in actual operation. Furthermore, the constraint on spindle speed is that it cannot exceed a preset percentage segment of the rated speed, such as 60% to 90% of the rated speed; the constraint on feed rate is that it cannot exceed the equipment's maximum feed rate; and the constraint on tool life allowance is that it must be greater than zero. Based on these constraints, a valid sample set is determined. Further, by simulating different combinations of spindle speed, feed rate, and tool life allowance, samples that meet all constraints are selected. Through these constraints, a valid sample set is selected from a large number of simulated samples.

[0052] The effective sample set is non-dominated and sorted to obtain a non-dominated solution set. Specifically, the effective sample set is divided into different levels using a non-dominated sorting algorithm. Samples in each level are not superior to samples in other levels in terms of both lifetime consumption and processing accuracy. The resulting non-dominated solution set contains solutions that achieve the optimal balance between lifetime consumption and processing accuracy. Through the above steps, non-dominated solutions are selected from the effective samples using non-dominated sorting. These non-dominated solutions achieve the optimal balance between lifetime consumption and processing accuracy. This constraint-based and non-dominated sorting-based optimization method can significantly improve equipment operating efficiency and production stability, reduce equipment failures and low production efficiency caused by improper parameter settings, and achieve optimized configuration of processing parameters.

[0053] Furthermore, by deploying edge monitoring nodes to collect tool wear signals, the method of this application includes: A wavelet transform unit is integrated on the edge monitoring node to obtain a subset of degradation-sensitive features. The subset of degradation-sensitive features is fused with the cutting parameters in the process execution information and used as the input of the gated loop unit network to predict the wear change trend within the production cycle of the process execution information. The normal wear range is then combined to determine whether a tool change command is triggered.

[0054] Specifically, wavelet transform units decompose signals into components of different frequencies and extract time-frequency features from the signals. Specifically, they are used to extract degradation-sensitive features from tool wear signals. A subset of degradation-sensitive features refers to a set of features strongly correlated with tool degradation extracted from the tool wear signal, thus reflecting early signs of tool wear. Gated recurrent unit networks are variants of recurrent neural networks that, by processing time-series data, can capture long-term dependencies in the sequence and are used to predict the changing trend of tool wear. Cutting parameters refer to parameters related to the cutting process in the process execution information, including cutting speed, feed rate, and depth of cut. These parameters directly affect tool wear. Normal wear range refers to the allowable wear range of the tool under normal machining conditions; tools exceeding the normal wear range need to be replaced.

[0055] Execution steps: A wavelet transform unit is integrated on the edge monitoring node to process the acquired tool wear signal. The tool wear signal is a time series data. It is decomposed into multiple frequency band feature components through wavelet transform. Energy entropy and kurtosis features are extracted from these feature components to generate time-frequency domain sensitive feature vectors. A subset of degradation-sensitive features that are strongly correlated with tool degradation is selected. Among the feature components extracted by wavelet transform, the increase of high-frequency components and the change of energy entropy may be related to early signs of tool wear. These features are selected as the degradation-sensitive feature subset.

[0056] The process integrates a subset of degradation-sensitive features with cutting parameters from the process execution information. Furthermore, it combines this subset with cutting parameters such as cutting speed, feed rate, and depth of cut to form a comprehensive feature vector. This comprehensive feature vector serves as the input to a gated loop unit (GRU) network to predict the trend of tool wear. Specifically, the GRU network predicts the trend of tool wear within the next production cycle; this is combined with the normal wear range to determine whether a tool change command should be triggered. In these steps, wavelet transform and GRU network prediction predict the trend of tool wear and issue a change command before the tool wear exceeds the normal range. This tool wear prediction method based on wavelet transform and GRU network significantly improves the efficiency and accuracy of tool management, reduces machining quality problems and equipment failures caused by tool wear, achieves real-time monitoring and prediction of tool wear, and issues tool change commands promptly based on the prediction results.

[0057] Furthermore, by integrating wavelet transform units on the edge monitoring nodes to obtain a subset of degradation-sensitive features, the method of this application includes: The tool wear signal is decomposed into feature components of multiple frequency bands; energy entropy and kurtosis features are extracted from the feature components of the multiple frequency bands to generate time-frequency domain sensitive feature vectors, and a subset of degradation sensitive features that are strongly correlated with tool degradation is selected.

[0058] Specifically, the characteristic components of a frequency band refer to the components in different frequency ranges decomposed into a signal through wavelet transform. Each characteristic component of a frequency band represents the characteristics of the signal within that frequency range. Energy entropy measures the uniformity of the signal's energy distribution; the higher the energy entropy, the more uniform the energy distribution of the signal. The lower the energy entropy, the more concentrated the signal's energy is in certain specific frequency components. Kurtosis measures the intensity of spikes and pulses in a signal; the higher the kurtosis, the more pronounced the spikes and pulses in the signal, usually associated with signal abrupt changes or anomalies. Time-frequency domain sensitive feature vectors are feature vectors that combine time and frequency information, reflecting the characteristics of the signal at different times and frequencies, and are used for subsequent feature analysis and classification. Deterioration-sensitive feature subsets refer to features strongly correlated with tool deterioration selected from the time-frequency domain sensitive feature vectors. These features can effectively reflect early signs of tool wear.

[0059] Execution steps: Decompose the tool wear signal into feature components of multiple frequency bands. For example, if the tool wear signal is a time series data, it can be decomposed into feature components of low frequency, mid frequency, and high frequency bands through wavelet transform. The feature component of each frequency band represents the characteristics of the signal within that frequency range. The low frequency component usually reflects the long-term trend of the signal, while the high frequency component reflects the short-term fluctuations and abrupt changes of the signal. Extract energy entropy and kurtosis features from the feature components of these frequency bands to generate time-frequency domain sensitive feature vectors. For example, calculate the energy entropy and kurtosis of each feature component of the frequency band, and combine these feature values ​​of energy entropy and kurtosis of the low frequency component, energy entropy and kurtosis of the mid frequency component, and energy entropy and kurtosis of the high frequency component into a time-frequency domain sensitive feature vector.

[0060] A subset of degradation-sensitive features strongly correlated with tool wear is selected. Specifically, by analyzing historical data and expert knowledge, it is determined that the kurtosis of high-frequency components is strongly correlated with early signs of tool wear. Therefore, kurtosis features of high-frequency components are selected from the time-frequency domain sensitive feature vector to form a subset of degradation-sensitive features. Through the above steps, early signs of tool wear are identified using wavelet transform and feature extraction, and timely warnings are issued before tool wear exceeds the normal range. This wavelet transform and feature extraction-based method can significantly improve the efficiency and accuracy of tool wear monitoring, reduce machining quality problems and equipment failures caused by tool wear, improve production stability, and extract features strongly correlated with tool wear from complex tool wear signals, providing an important basis for subsequent fault prediction and maintenance.

[0061] In summary, the beneficial effects of the embodiments of this application are: This application provides a method and system for fault monitoring of precision component production equipment based on digital twins. It collects equipment status data, including spindle vibration amplitude and tool temperature field, under the same time stamp, based on sensor monitoring data and process execution information. A digital twin model of the equipment is constructed using sensor monitoring data and a fault mode knowledge base. Multi-dimensional simulation and deduction are performed using the equipment status data to determine early warning thresholds and fault response strategies for key components. Edge monitoring nodes are deployed to collect tool wear signals. Combined with the machining accuracy decay trend within the production cycle of process execution information, wear anomaly intervals are matched using cosine similarity. Deterioration-sensitive features under an attention mechanism are set to generate a fault evolution prediction map within a time sliding window, and adaptive early warning level matching and control command compensation are performed. Simultaneously, a fault propagation network is constructed based on the early warning thresholds and fault response strategies for key components to mark and remind users of the fault impact range. This application achieves the technical effect of collecting tool wear signals, matching anomaly intervals using cosine similarity, generating a fault evolution prediction map, improving the accuracy and coordination of fault monitoring, and providing reliable assurance for the production stability of precision components.

[0062] Example 2, based on the same inventive concept as the fault monitoring method for precision component manufacturing equipment based on digital twins in the previous examples, such as... Figure 2 As shown in the figure, this application provides a fault monitoring system for precision component manufacturing equipment based on digital twins, wherein the system includes: Data collection module M100: Based on sensor monitoring data and process execution information, it collects equipment status data, including spindle vibration amplitude and tool temperature field, under the same timestamp.

[0063] Simulation and deduction module M200: Based on the sensor monitoring data and fault mode knowledge base, it constructs a digital twin model of the equipment and performs multi-dimensional simulation and deduction in conjunction with the equipment status data to determine the early warning thresholds of key components and fault response strategies.

[0064] Fault Evolution Module M300: Deploys edge monitoring nodes to collect tool wear signals, combines the machining accuracy decay trend within the production cycle of the process execution information, matches abnormal wear intervals through cosine similarity, sets degradation-sensitive features under the attention mechanism, generates a fault evolution prediction map within a time sliding window, and performs adaptive early warning level matching and control command compensation.

[0065] Fault marking and reminder module M400: Simultaneously, based on the warning threshold of the key components and the fault response strategy, a fault propagation network is constructed to mark and remind the scope of fault impact.

[0066] Furthermore, the fault marker reminder module M400 is also used to perform the following method: Based on the fault propagation network, key fault indicators, including bearing temperature rise rate and guide rail clearance change rate, are extracted; based on the key fault indicators, combined with equipment operating conditions and process accuracy requirements, dynamic scheduling parameters and priority response rules for maintenance resources are configured.

[0067] Furthermore, the fault marker reminder module M400 is also used to perform the following method: The maintenance resources interact with the digital twin model of the equipment in real time; the scheduling path of the maintenance resources is dynamically adapted to the equipment downtime window, and the fault mode knowledge base is iteratively updated using a deep reinforcement learning algorithm.

[0068] Furthermore, the fault marker reminder module M400 is also used to perform the following method: A two-dimensional reward function is set up, where the first dimension is the matching degree between the fault prediction deviation rate and the actual fault degree, and the second dimension is the maintenance response time. Based on the two-dimensional reward function, a deep Q-network is used to learn the fault identification strategy in the discrete fault type space.

[0069] Furthermore, the data collection module M100 is used to perform the following methods: The spindle vibration amplitude and tool temperature field are bound to the production process nodes and processing batch information to establish a state-process correlation matrix. Based on the state-process correlation matrix, the state differences of the same production equipment at different processing stages are identified, and the baseline state feature values ​​are extracted. The baseline state feature values ​​are used to automatically identify abnormal processes and push them to the equipment health management terminal.

[0070] Furthermore, the simulation and deduction module M200 is used to execute the following methods: A material fatigue curve library for precision components is embedded in the digital twin model of the equipment. The material fatigue curve library contains life decay parameters under different stress amplitudes. Based on the material fatigue curve library, the stress accumulation process of key components under different processing loads and different cutting path combinations is simulated to generate the life consumption Pareto front and the processing accuracy Pareto front. According to the production accuracy priority, the optimal equilibrium point in the non-dominated solution set is selected from the life consumption Pareto front and the processing accuracy Pareto front.

[0071] Furthermore, the simulation and deduction module M200 is also used to perform the following methods: Spindle speed, feed rate, and tool life allowance are used as variable parameters, and constraints are set in the digital twin model of the equipment. Based on the constraints, a valid sample set is determined, and non-dominated solution set is obtained by non-dominated sorting.

[0072] Furthermore, the fault evolution module M300 is used to execute the following method: A wavelet transform unit is integrated on the edge monitoring node to obtain a subset of degradation-sensitive features. The subset of degradation-sensitive features is fused with the cutting parameters in the process execution information and used as the input of the gated loop unit network to predict the wear change trend within the production cycle of the process execution information. The normal wear range is then combined to determine whether a tool change command is triggered.

[0073] Furthermore, the fault evolution module M300 is also used to perform the following methods: The tool wear signal is decomposed into feature components of multiple frequency bands; energy entropy and kurtosis features are extracted from the feature components of the multiple frequency bands to generate time-frequency domain sensitive feature vectors, and a subset of degradation sensitive features that are strongly correlated with tool degradation is selected.

[0074] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0075] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. A method for monitoring the failure of a precision parts production plant based on digital twinning, characterized in that, The method comprises: Based on the sensing monitoring data, process execution information, collect equipment state data including spindle vibration amplitude, tool temperature field under the same timestamp limit; According to the sensing monitoring data, fault mode knowledge base, construct a device digital twin model, and combine the equipment state data to carry out multi-dimensional simulation deduction, determine the key component warning threshold and fault response strategy; Deploy edge monitoring nodes to collect tool wear signal, combine the processing precision attenuation trend in the production beat of the process execution information, match the wear abnormal interval through cosine similarity, set the degradation sensitive features under the attention mechanism, generate the fault evolution prediction graph in the time sliding window, and carry out adaptive warning level matching and control instruction compensation; At the same time, based on the key component warning threshold and fault response strategy, a fault conduction network is constructed to mark and remind the fault influence range.

2. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 1, wherein, Based on the key component warning threshold and fault response strategy, a fault conduction network is constructed to mark and remind the fault influence range, the method further comprises: According to the fault conduction network, extract key fault indicators including bearing temperature rise rate and guideway gap change rate; According to the key fault indicators, combine the equipment operation condition and process precision requirement to configure the dynamic scheduling parameters and priority response rules of maintenance resources.

3. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 2, wherein, The maintenance resources interact with the device digital twin model in real time; The scheduling path of the maintenance resources is dynamically adapted to the equipment downtime window, and a deep reinforcement learning algorithm is used to iteratively update the fault mode knowledge base.

4. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 3, wherein, Using a deep reinforcement learning algorithm to iteratively update the fault mode knowledge base, the method comprises: Set a two-dimensional reward function, wherein the first dimension is the matching degree of fault prediction deviation rate and actual fault degree, and the second dimension is the maintenance response time; Based on the two-dimensional reward function, a deep Q network is used to learn fault identification strategy in discrete fault type space.

5. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 1, wherein, Collecting equipment state data including spindle vibration amplitude and tool temperature field under the same timestamp limit, the method comprises: Bind the spindle vibration amplitude and tool temperature field with the production process node and processing batch information to establish a state-process association matrix; Based on the state-process association matrix, identify the state difference of the same production equipment in different processing stages, and extract the reference state characteristic value; Use the reference state characteristic value to automatically identify state abnormal processes and push to the equipment health management terminal.

6. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 5, wherein, Constructing a device digital twin model and combining the equipment state data to carry out multi-dimensional simulation deduction to determine the key component warning threshold and fault response strategy, the method comprises: Embed the material fatigue curve library of precision components in the device digital twin model, and the material fatigue curve library contains life attenuation parameters under different stress amplitudes; According to the material fatigue curve library, simulate the stress accumulation process of key components under different processing loads and different cutting path combinations to generate life consumption Pareto frontier and machining precision Pareto frontier; According to the production precision priority, an optimal balance point in the non-dominated solution set is selected from the life consumption Pareto frontier and the machining precision Pareto frontier.

7. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 6, wherein, The method comprises: The spindle speed, the feed speed, and the tool life margin are taken as variable parameters, and constraint conditions are set in the equipment digital twin model; Based on the constraint conditions, an effective sample set is determined, and non-dominated sorting is performed to obtain the non-dominated solution set.

8. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 1, wherein, An edge monitoring node is deployed to collect the tool wear signal, and the method comprises: A wavelet transform unit is integrated on the edge monitoring node to obtain a degradation sensitive feature subset; The degradation sensitive feature subset is fused with the cutting parameters in the process execution information as the input of a gated recurrent unit network to predict the wear trend within the production tempo of the process execution information, and whether to trigger a tool replacement instruction is determined in combination with the normal wear range.

9. The digital-twin-based precision-part-production-equipment-failure-monitoring method of claim 8, wherein, A wavelet transform unit is integrated on the edge monitoring node to obtain a degradation sensitive feature subset, and the method comprises: The tool wear signal is decomposed into feature components of multiple frequency bands; Energy entropy and kurtosis features are extracted from the feature components of the multiple frequency bands to generate a time-frequency domain sensitive feature vector, and a degradation sensitive feature subset strongly related to tool degradation is screened.

10. A precision parts production equipment failure monitoring system based on digital twinning, characterized in that, The system for implementing the steps of the digital twin-based precision part production equipment fault monitoring method according to any one of claims 1-9, the system comprising: A data collection module: based on the sensing monitoring data and the process execution information, collecting equipment state data including spindle vibration amplitude and tool temperature field under the same timestamp limitation; An simulation deduction module: constructing an equipment digital twin model according to the sensing monitoring data and the fault mode knowledge base, and performing multi-dimensional simulation deduction combined with the equipment state data to determine a key component warning threshold and a fault response strategy; A fault evolution module: deploying an edge monitoring node to collect a tool wear signal, combining the machining precision attenuation trend within the production tempo of the process execution information, matching the wear abnormal interval through cosine similarity, setting degradation sensitive features under the attention mechanism, generating a fault evolution prediction graph within a time sliding window, and performing adaptive warning level matching and control instruction compensation; A fault marking reminder module: meanwhile, constructing a fault conduction network based on the key component warning threshold and the fault response strategy to mark and remind the fault influence range.

Images