Intelligent fault diagnosis method and system for automatic numerical control machine tool

Abstract

The invention discloses an intelligent fault diagnosis method and system for an automatic numerical control machine tool, and relates to the technical field of data processing, and the method comprises the steps: interacting fault event records of a target numerical control machine tool; based on the fault event records, fault characteristic mining is carried out to construct a fault network, a fault characteristic network is constructed through graph theory analysis based on fault propagation, and a fault probability network is constructed through fault failure analysis; according to the fault network, diagnosis large model training and lightweight processing are carried out, a fault diagnosis module is built, and module building is carried out through two-step training based on a lightweight loss function; time-staggered sampling frequency is determined, monitoring management and control of a front-end sensor are carried out based on the time-staggered sampling frequency, and machine tool operation data are determined; and the machine tool operation data is transmitted back to the fault diagnosis module, fault occurrence judgment and failure level measurement are carried out, and a fault diagnosis result is determined. Therefore, the technical effects of improving the fault diagnosis accuracy and real-time performance and improving the deployment adaptability are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an intelligent fault diagnosis method and system for an automated numerically controlled machine tool. Background Art

[0002] Various faults are inevitable in the operation of CNC machine tools. Especially when they are operated for a long time at high load, the faults may have a serious impact on production efficiency and product quality, and even cause economic losses to the enterprise.

[0003] The fault propagation of CNC machine tools has complex spatiotemporal characteristics, and traditional diagnostic methods are difficult to effectively mine the correlation between faults. At the same time, due to the complex operating environment of CNC machine tools and inconsistent sensor data collection frequencies, traditional methods are difficult to efficiently integrate multi-source heterogeneous data. In addition, most of the existing diagnostic models are general models that lack lightweight processing and are difficult to meet the deployment requirements of edge devices. In summary, the existing technologies have technical problems such as limited integration capabilities of multi-source heterogeneous data, insufficient correlation mining, and insufficient real-time and lightweight diagnostic models. Summary of the invention

[0004] The present invention provides an intelligent fault diagnosis method and system for automated CNC machine tools to solve the technical problems in the prior art of limited integration capability of multi-source heterogeneous data, insufficient association mining, and insufficient real-time and lightweight of the diagnosis model, thereby achieving the technical effect of improving the accuracy and real-time performance of fault diagnosis and improving deployment adaptability.

[0005] In a first aspect, the present invention provides an intelligent fault diagnosis method for an automated numerically controlled machine tool, wherein the method comprises: Interactive target CNC machine tool fault event record.

[0006] Based on the fault event records, fault characteristics mining is performed to construct a fault network, wherein the fault feature network is constructed by performing graph theory analysis based on fault propagation, and the fault probability network is constructed by performing fault failure analysis.

[0007] According to the fault network, a large diagnostic model is trained and lightweight processed to build a fault diagnosis module, wherein the module is built by two-step training based on a lightweight loss function.

[0008] Determine the staggered sampling frequency, monitor and control the front-end sensor based on the staggered sampling frequency, and determine the machine tool operation data.

[0009] The machine tool operation data is transmitted back to the fault diagnosis module to perform fault occurrence determination and failure level measurement to determine the fault diagnosis result.

[0010] In a second aspect, the present invention further provides an intelligent fault diagnosis system for an automated numerically controlled machine tool, wherein the system comprises: The CNC machine tool fault event interaction component is used to interact with the target CNC machine tool fault event records.

[0011] The fault characteristic mining and fault network construction component is used to mine fault characteristics and construct a fault network based on the fault event record, wherein the fault characteristic network is constructed by performing graph theory analysis based on fault propagation, and the fault probability network is constructed by performing fault failure analysis.

[0012] A fault diagnosis module building component is used to perform large-scale diagnostic model training and lightweight processing according to the fault network, and build a fault diagnosis module, wherein the module is built by two-step training based on a lightweight loss function.

[0013] The front-end sensor monitoring and control component is used to determine the staggered sampling frequency, monitor and control the front-end sensor based on the staggered sampling frequency, and determine the machine tool operation data.

[0014] The fault diagnosis execution component is used to transmit the machine tool operation data back to the fault diagnosis module, perform fault occurrence judgment and failure level measurement, and determine the fault diagnosis result.

[0015] The present invention discloses an intelligent fault diagnosis method and system for an automated numerically controlled machine tool, comprising: recording interactive target fault events of the numerically controlled machine tool; mining fault characteristics to construct a fault network, comprising: constructing a fault feature network based on fault propagation through graph theory analysis, and constructing a fault probability network through failure analysis; training and simplifying a large-scale diagnosis model based on the constructed fault network to establish a fault diagnosis system, wherein a lightweight loss function is used to perform two-stage training to construct the system; setting an asynchronous sampling frequency, monitoring and managing front-end sensors according to the frequency to obtain machine tool operation data; sending the machine tool operation data back to the fault diagnosis system to perform fault judgment and failure level evaluation to determine the final fault diagnosis result. The intelligent fault diagnosis method and system for an automated numerically controlled machine tool disclosed by the present invention solve the technical problems of limited integration capability of multi-source heterogeneous data, insufficient correlation mining, and insufficient real-time and lightweight of the diagnosis model, and achieve the technical effects of improving the accuracy and real-time performance of fault diagnosis and improving deployment adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a flow chart of the intelligent fault diagnosis method for automated numerically controlled machine tools of the present invention.

[0017] Figure 2 It is a structural schematic diagram of the intelligent fault diagnosis system of the automated numerically controlled machine tool of the present invention.

[0018] Explanation of the accompanying drawings: CNC machine tool fault event interaction component 11, fault characteristic mining and fault network construction component 12, fault diagnosis module construction component 13, front-end sensor monitoring and control component 14, fault diagnosis execution component 15. DETAILED DESCRIPTION

[0019] The technical solution provided in the embodiments of the present invention is to solve the technical problems of the prior art, such as limited integration capability of multi-source heterogeneous data, insufficient association mining, and insufficient real-time and lightweight of the diagnostic model. The overall idea adopted is as follows: First, the fault event records of the interactive target CNC machine tool are collected, and the fault characteristics are mined based on these fault event records to build a fault network. Specifically, a fault feature network is constructed through graph theory analysis based on fault propagation, and a fault probability network is constructed through fault failure analysis. Then, the fault network is used to train and lightweight the large diagnostic model to build a fault diagnosis module. In this process, a two-step training method based on a lightweight loss function is used to build the module. Then, the staggered sampling frequency is determined, and the front-end sensors are monitored and controlled based on the frequency to ensure the collection of machine tool operation data. Finally, the machine tool operation data is transmitted back to the fault diagnosis module, and the fault diagnosis result is determined by fault occurrence judgment and failure level measurement.

[0020] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0021] Embodiment 1 Figure 1 The figure is a flow chart of the intelligent fault diagnosis method of the automatic numerically controlled machine tool of the present invention, wherein the method comprises: S100: Interactive target CNC machine tool fault event record.

[0022] Specifically, the target CNC machine tool fault event record is the historical fault log of the target CNC machine tool, which is sample data for fault diagnosis. Exemplarily, the target CNC machine tool fault event record includes fault time, fault description, fault level, equipment metadata (such as the equipment's operating temperature, working pressure, speed, current, vibration frequency, position deviation), etc.

[0023] S200: Based on the fault event record, fault characteristic mining is performed to construct a fault network, wherein the fault characteristic network is constructed by performing graph theory analysis based on fault propagation, and the fault probability network is constructed by performing fault failure analysis.

[0024] Specifically, by mining fault characteristics through fault event records, building fault feature networks and fault probability networks, we can reveal the relationships between different faults, fault propagation paths, and the possibility of fault occurrence.

[0025] Specifically, methods based on graph theory analysis and failure analysis can comprehensively evaluate fault risks, predict equipment failure modes, and generate fault networks, providing a scientific basis for equipment maintenance decisions. Graph theory analysis is used to explore the paths and impact relationships of fault propagation to generate and construct a fault feature network. Based on fault history data, the failure probability of each fault component is analyzed, and probabilistic network modeling is performed to obtain and construct a fault probability network.

[0026] In some embodiments, based on the fault event record, fault characteristic mining is performed, including: Acquire a first fault event, determine fault intrinsic data and fault propagation data; perform multi-layer segmentation and feature extraction on the fault intrinsic data, perform hierarchical compensation based on the fault propagation data, and determine a first event segmentation tree; perform node feature filtering on the first event segmentation tree by setting a preset correlation, and determine a first valid segmentation tree; and identify a probability threshold of node failure on the first valid segmentation tree.

[0027] Specifically, first, extract relevant data of the first fault event from the fault event record. The first fault event refers to a fault record that occurs at any time. The first fault event includes information such as the time when the fault occurred, the fault type, the fault component, the fault cause, and the repair time; then, extract key information from the fault event record, including the environmental conditions under which the fault occurred, the equipment operating status, etc., and determine it as the fault intrinsic data. In other words, the fault intrinsic data refers to the original data directly related to the fault; then, use causal reasoning analysis tools to explore the causal relationship between different faults, establish a fault propagation model, and determine the fault propagation data from the fault event record. The fault propagation data includes the impact of the fault on other components after the fault occurs. Fault propagation analysis helps to reveal how a fault triggers subsequent other faults and build a fault propagation chain.

[0028] Specifically, the fault intrinsic data is processed and analyzed in layers, and the fault events are decomposed into multiple levels or dimensions to facilitate in-depth understanding of the characteristics of each level, where each level can represent different states of the equipment or characteristics of different dimensions; illustratively, the fault data is classified by level through a hierarchical clustering analysis method to reveal the relationship between its different levels; then, in each level, representative and discriminative feature data is extracted, for example, characteristics such as the frequency, duration, and repair cycle of the fault, as well as parameters such as the fault type and equipment load can be extracted.

[0029] Specifically, based on the fault propagation data, analyze how the fault propagates from one component to other components, and perform compensation processing on the data in the propagation process, exemplarily including interpolating or extrapolating missing data or abnormal values ​​in time to ensure the continuity and accuracy of the data, and interpolating or extrapolating missing data or abnormal values ​​in time to ensure the continuity and accuracy of the data; through the above process, the nodes of the constructed first event segmentation tree represent different characteristics or states of the fault, and the edges of the tree represent the propagation path of the fault from one state to another.

[0030] Furthermore, according to the preset correlation threshold, the relationship strength between each node and other nodes is evaluated. The more relevant the node features are, the more likely they are to become a key factor affecting fault propagation. If the relationship strength between a node and other nodes is lower than the correlation threshold, the corresponding fault propagation relationship has a lower reliability and can be eliminated. The first effective segmentation tree obtained after this step will retain important fault information and remove irrelevant information.

[0031] Exemplarily, the Pearson correlation coefficient is calculated to quantify the correlation between different features, and then the mutual information between nodes is calculated to determine their importance in fault propagation.

[0032] Optionally, a failure probability is determined for each leaf node based on a statistical analysis method (such as failure mode effect analysis, Bayesian inference model, reliability analysis), indicating the possibility of failure of the node under different conditions.

[0033] The above method steps, through multi-level segmentation and node feature filtering, can effectively improve the accuracy of the fault prediction model and provide accurate data support for the preventive maintenance and fault diagnosis of equipment; at the same time, based on the identification of effective segmentation trees and failure probabilities, it can optimize maintenance strategies, concentrate resources on key fault nodes, and improve the reliability and stability of equipment.

[0034] S300: According to the fault network, a large diagnostic model is trained and lightweight processed to build a fault diagnosis module, wherein the module is built by two-step training based on a lightweight loss function.

[0035] Specifically, by utilizing the information in the fault network and combining it with lightweight model training technology, an efficient and accurate fault diagnosis module can be constructed to take into account both the high accuracy of the model and the lightweight requirements of rapid response.

[0036] In some embodiments, the performing of diagnostic large model training comprises: The fault event records are sorted to construct fault diagnosis samples, wherein the fault diagnosis samples include fault discrimination samples and failure degree analysis samples; based on the fault feature network as a benchmark, supervised training is performed based on the fault discrimination samples to obtain a diagnosis decision block; based on the fault probability network as a benchmark, supervised training is performed based on the failure degree analysis samples to obtain a failure decision block; the failure decision block is post-connected to the diagnosis decision block to determine a large diagnosis model.

[0037] Specifically, to train the large diagnosis model, first, the historical fault event records are structured and organized to facilitate subsequent analysis and training, including deduplication, correction and supplementation of historical fault event records, and marking each fault event record with corresponding fault type, occurrence time, affected equipment, fault cause and other information.

[0038] Specifically, fault diagnosis samples include constructing fault identification samples and failure analysis samples. Fault identification samples are composed of fault features (such as equipment sensor data, historical failure modes, etc.) and fault labels (such as fault type, whether the fault occurs, etc.), which are used to train the fault diagnosis module to determine whether the equipment fails and the type of fault; failure analysis samples contain equipment operating status characteristics (such as equipment parameters, performance changes, etc.) and failure probability labels (such as failure risk assessment, expected failure time, etc.), which are used to train the failure analysis module, mainly to determine the possibility of equipment failure caused by faults.

[0039] Specifically, a diagnostic decision block based on deep learning (such as multi-layer perceptron, convolutional neural network, graph convolutional network, etc.) is constructed, and the sorted fault identification samples are used in combination with the fault feature network for supervised training. The diagnostic decision block optimizes the model parameters by learning the mapping relationship between input data (equipment operating status, sensor data, etc.) and target output (fault type or whether a fault occurs), thereby acquiring the ability to predict faults. After training, the diagnostic decision block will be able to analyze the input fault feature data and output information such as fault type and whether a fault occurs. As the basic part of the entire diagnostic model, it is responsible for preliminary fault identification.

[0040] Specifically, based on the same method principle as that of obtaining the diagnosis decision block, a failure decision block is constructed and trained. The failure decision block is used to determine the possibility of equipment failure caused by the existing fault based on the decision of the diagnosis decision block. In other words, only when the diagnosis decision block determines that there is an abnormality, the failure decision block is enabled to perform failure analysis to obtain the degree of operational abnormality caused by the fault, while when the diagnosis is normal, there is no need to perform failure analysis.

[0041] Specifically, after training is completed, the failure decision block will be able to output the equipment's failure score, failure probability and other information based on the input equipment status data and fault probability network, providing support for failure prediction and risk assessment for the large diagnostic model.

[0042] By separately setting up and performing timing analysis on the diagnostic decision block and the failure decision block, it helps to avoid reactive operations under fusion analysis, i.e., performing failure analysis on data without faults, while improving the pertinence and accuracy of decisions.

[0043] In some embodiments, lightweight processing is performed to build a fault diagnosis module, including: Constructing a lightweight loss function, wherein the lightweight loss function represents the transmission loss from the large diagnostic model to the fault diagnosis module, and includes a logical dimension and a data dimension; The output of the large diagnosis model is used as the first sample, the fault diagnosis sample is used as the second sample, and transfer training is performed in combination with the lightweight loss function to determine the fault diagnosis module.

[0044] Specifically, in order to improve the processing efficiency of the fault diagnosis system, reduce the amount of calculation, and maintain a high diagnostic accuracy, the diagnostic model is lightweight; illustratively, first, a lightweight loss function is constructed, the goal of which is to measure the transmission loss from a complex large diagnostic model to a simplified fault diagnosis module; the loss function should consider the logical dimension and data dimension of the model to ensure a balance between accuracy and efficiency.

[0045] Specifically, the logical dimension of the loss function is used to measure the loss of reasoning ability of the simplified model in fault diagnosis. For example, the loss can be quantified by comparing the difference in reasoning paths (decision-making process) between the simplified model and the original model under the same input; the data dimension loss is used to measure the loss of the simplified model in data output, ensuring that the output of the simplified model can retain the original fault diagnosis data pattern as much as possible. The loss function can be calculated by comparing the difference in output data (such as fault type, failure probability, etc.) between the simplified model and the original model.

[0046] Optionally, the logical dimension loss and the data dimension loss are combined through a weighted coefficient to form a final lightweight loss function to balance the logical accuracy and data accuracy.

[0047] Specifically, before simplifying the fault diagnosis model, the output of the large diagnosis model is first used as the first sample, which is combined with the fault diagnosis sample and passed to the lightweight loss function for preliminary training; after the preliminary training, the simplified model has a certain fault discrimination ability, but there may still be some accuracy loss, so further refined training is carried out through fault diagnosis samples to improve its performance in practical applications.

[0048] Through the above steps, combined with a lightweight loss function and a two-step training process, the original large diagnostic model can be effectively simplified into an efficient fault diagnosis module, while maintaining the accuracy of the model and greatly improving the processing efficiency to meet the needs of real-time fault detection; in addition, the lightweight model also has good adaptability and scalability, and can be flexibly applied to different types of equipment and fault scenarios.

[0049] In some implementations, transfer training is performed in conjunction with the lightweight loss function, including: Obtain a first fault diagnosis sample, input the diagnostic large model, determine first diagnostic data and add it into the first sample, wherein the first diagnostic data includes the entire diagnostic process based on the diagnostic large model; set the learning rate for the first fault diagnosis sample and the first diagnostic data, and perform a lightweight training on the fault diagnosis module with the lightweight loss function as a constraint.

[0050] Specifically, first, a first fault diagnosis sample is randomly obtained and input into the diagnosis big model, and the output data of the diagnosis big model, i.e., the diagnosis intermediate data, is stored as the first diagnosis data and added into the first sample; in other words, the goal of the diagnosis big model is to provide a complete fault analysis based on the input data, including: determination of the fault type, analysis of the cause of the fault, prediction of the fault propagation path, analysis of the probability of occurrence and failure rate of the fault, and the generated first sample includes the complete decision-making process of the diagnosis big model, including how the model draws the diagnosis conclusion from the input data, to ensure the transparency and traceability of the diagnosis process.

[0051] Specifically, the fault diagnosis module is trained based on the first fault diagnosis sample and the first diagnostic data, and a suitable learning rate is set, wherein the learning rate refers to the step size of the model parameter update in each training iteration. Preferably, during lightweight training, a smaller initial learning rate can be selected to ensure the stability of the training process. At the same time, dynamically adjusting the learning rate based on the historical information of each gradient can more effectively adapt to the training needs of complex models.

[0052] Exemplarily, in each training iteration, first, the first fault diagnosis sample is input into the diagnosis module to calculate the output of the simplified model; then, the error of the simplified model is calculated according to the lightweight loss function, and the parameters of the fault diagnosis module are backpropagated based on the error to optimize its weights and parameters; then, the first fault diagnosis sample is replaced by the first diagnostic data for iterative training until the function value of the lightweight loss function meets the constraints. After lightweight training, the simplified model will reduce redundant calculations while retaining key information, so as to efficiently perform fault diagnosis.

[0053] Through the above steps, transfer training combined with lightweight loss functions can effectively realize the transformation of the large diagnostic model to a lightweight fault diagnosis module. It not only retains the key information and efficiency of the large diagnostic model, but also simplifies the calculation process and model structure, ensuring that the final model has high diagnostic accuracy and can run efficiently in a resource-constrained environment.

[0054] S400: Determine a staggered sampling frequency, monitor and control the front-end sensor based on the staggered sampling frequency, and determine the machine tool operation data.

[0055] In some embodiments, before determining the staggered sampling frequency, the process includes: The operating mechanism of the target CNC machine tool is obtained, and multiple groups of operating elements are determined by scene decomposition; for the multiple groups of operating elements, a simultaneous trigger mark of scene monitoring is performed to determine a first sampling rule; the multiple groups of operating elements are traversed, and based on the element fluctuation characteristics, the sampling frequency is set to determine a second sampling rule, wherein the element fluctuation is positively correlated with the sampling frequency; based on the first sampling rule and the second sampling rule, the front-end sensor is configured, wherein the monitoring sampling control of the front-end sensor is performed with multiple threads.

[0056] Specifically, in order to reasonably set the sampling frequency, it is necessary to have an in-depth understanding of the state changes of the machine tool under different working conditions. The analysis process includes a comprehensive study of the machine tool's control system, drive system, operating conditions and the dynamic behavior of various sensors to clarify its key performance indicators; among them, the operating mechanism refers to the operation process of the CNC machine tool under various working modes, including the machine tool's dynamics, control system, sensor response, etc.

[0057] Specifically, scenario decomposition refers to decomposing the entire operation process of the machine tool into multiple sub-scenarios, each sub-scenario representing the operating characteristics of the machine tool in a certain state. Exemplarily, the scenarios include: idle state, processing state, start / stop state, fault state, etc.

[0058] Specifically, operating factors refer to the key parameter dimensions of the machine tool in various scenarios, such as speed, temperature, pressure, vibration, load, etc. The operating factors can reflect the operating status of the machine tool. By analyzing the changes in the operating factors, faults can be predicted or the performance of the machine tool can be analyzed. Among them, after each scenario is decomposed, it may involve multiple different operating factors, that is, corresponding to multiple groups of operating factors.

[0059] Specifically, during the operation of the machine tool, different scenarios may occur simultaneously. For example, when the machine tool switches from an idle state to a processing state, the parameters of some sensors may change suddenly, or the data changes between different sensors may have a certain time delay. The simultaneous trigger flag is used to determine the trigger timing relationship of multiple sensor data to ensure that in different scenarios, related sensors can be activated simultaneously within a reasonable time window for data collection and analysis.

[0060] Specifically, the first sampling rule is determined based on the synchronization requirement between multiple sensor data to ensure joint sampling of simultaneously triggered data, for example, activating the sensor in advance or increasing the acquisition frequency in advance when the sub-scene is switched.

[0061] Specifically, factor fluctuation characteristics refer to the changing characteristics of various operating factors of machine tools (such as vibration, temperature, load, etc.) in different time periods. The sensor frequency of the machine tool needs to adapt to these fluctuation characteristics to capture the periodicity and fluctuation laws of the changes. For example, for each operating factor, a different sampling frequency is set according to its fluctuation characteristics. If a certain factor fluctuates greatly, the sampling frequency should be appropriately increased; conversely, the sampling frequency of factors with smaller fluctuations can be appropriately reduced.

[0062] Specifically, the second sampling rule is a further optimization rule based on the fluctuation characteristics of the factors. Preferably, the second sampling rule includes not only the setting of the sampling frequency, but also the real-time adjustment of the sampling frequency, that is, during the operation of the machine tool, the sampling frequency is dynamically adjusted according to the real-time monitored fluctuations of the factors; for example, when the fluctuation of a certain factor exceeds the set threshold, the sampling frequency is increased; when the fluctuation of the factor is stable, the sampling frequency is reduced.

[0063] Through the above steps, the determination of staggered sampling frequency and the monitoring and control of front-end sensors can improve the efficiency and accuracy of machine tool fault monitoring; by optimizing the sampling frequency, unnecessary data redundancy is reduced, making subsequent data processing and analysis more efficient, reducing the computing pressure of the system, ensuring efficient data collection of the machine tool under different working conditions, and providing reliable support for the machine tool fault diagnosis system.

[0064] S500: The machine tool operation data is transmitted back to the fault diagnosis module to perform fault occurrence determination and failure level measurement to determine the fault diagnosis result.

[0065] Specifically, the machine tool operation data is used as the diagnostic input, and the fault diagnosis module performs fault diagnosis to determine whether a fault occurs (fault occurrence determination) and the fault level (failure level measurement), and provides fault diagnosis results to assist decision-making.

[0066] In some embodiments, the performing of fault occurrence determination and failure level measurement includes: Preprocess the machine tool operation data and transmit it to the fault diagnosis module, wherein data screening, dimensionality reduction and feature extraction are used as preprocessing standards; based on the diagnosis decision block of the fault diagnosis module, diagnose the occurrence of the fault and determine the diagnosis result of the fault event; if the diagnosis result is empty, terminate the diagnosis process; if the diagnosis result is not empty, transfer the diagnosis result to the failure decision block, determine the fault failure level, and output the fault diagnosis result.

[0067] Specifically, first, the machine tool operation data is preprocessed, including data screening, data dimensionality reduction and feature extraction, so as to eliminate redundant information, reduce data noise and improve data quality; then, in the fault diagnosis module, the diagnosis decision block is used to make a judgment on the occurrence of a fault based on the processed machine tool operation data to determine whether the machine tool has a fault. If the machine tool has not failed, an empty result is returned or marked as "no fault" and the diagnosis of the current machine tool operation data is terminated; if the machine tool fails, the diagnostic result including the fault type and related information (such as fault location, fault mode, etc.) is output, and the next step of failure level measurement is entered; then, the diagnostic result of the diagnosed fault is further analyzed through the failure decision block to evaluate the severity and impact range of the fault.

[0068] Specifically, the failure decision block analyzes the potential impact of the failure, the impact on the production process, and the degree of damage to the equipment, and evaluates the failure level of the machine tool after the failure occurs. Common failure levels include: minor failures, which have little impact on the performance of the machine tool and may require maintenance, but do not affect production; moderate failures, which have a greater impact on the performance of the machine tool and may lead to reduced production efficiency; severe failures, which cause the machine tool to stop working and may require immediate repair.

[0069] Specifically, the final fault diagnosis result includes two main contents: the fault type including the type, location and cause of the fault and the failure level reflecting the severity of the fault.

[0070] Through the above steps, the diagnosis decision block quickly determines whether the machine tool has a fault, and the failure decision block evaluates the failure level based on the diagnosis results to help operators intervene in time and reduce downtime, thereby improving the response speed and accuracy of machine tool failures, reducing maintenance costs, optimizing production processes, and providing effective decision support for machine tool maintenance and management.

[0071] In summary, the intelligent fault diagnosis method for automated CNC machine tools provided by the present invention has the following technical effects: By recording the interactive target fault events of CNC machine tools; mining fault characteristics to build a fault network, including: building a fault feature network based on fault propagation through graph theory analysis, and building a fault probability network through failure analysis; based on the constructed fault network, training and simplifying large-scale diagnostic models to establish a fault diagnosis system, in which a lightweight loss function is used to conduct two-stage training to build the system; setting an asynchronous sampling frequency, and monitoring and managing front-end sensors according to the frequency to obtain the operation data of the machine tools; sending these machine tool operation data back to the fault diagnosis system for fault judgment and failure level evaluation to determine the final fault diagnosis result, thereby achieving the technical effect of improving the accuracy and real-time performance of fault diagnosis and improving deployment adaptability.

[0072] Embodiment 2 Figure 2 Schematic diagram of the structure of the intelligent fault diagnosis system of the automatic numerical control machine tool of the present invention. Figure 1 The flow chart of the intelligent fault diagnosis method of the automatic numerical control machine tool of the present invention can be shown as follows: Figure 2 The structure shown is implemented.

[0073] Based on the same concept as the intelligent fault diagnosis method of the automatic numerically controlled machine tool in the above embodiment, the intelligent fault diagnosis system of the automatic numerically controlled machine tool further provided by the present invention comprises: The CNC machine tool fault event interaction component 11 is used for interacting with the target CNC machine tool fault event record.

[0074] The fault characteristic mining and fault network construction component 12 is used to mine fault characteristics and construct a fault network based on the fault event record, wherein the fault characteristic network is constructed by performing graph theory analysis based on fault propagation, and the fault probability network is constructed by performing fault failure analysis.

[0075] The fault diagnosis module building component 13 is used to perform large-scale diagnosis model training and lightweight processing according to the fault network to build a fault diagnosis module, wherein the module is built by two-step training based on a lightweight loss function.

[0076] The front-end sensor monitoring and control component 14 is used to determine the staggered sampling frequency, monitor and control the front-end sensor based on the staggered sampling frequency, and determine the machine tool operation data.

[0077] The fault diagnosis execution component 15 is used to transmit the machine tool operation data back to the fault diagnosis module, perform fault occurrence determination and failure level measurement, and determine the fault diagnosis result.

[0078] Among them, the fault characteristic mining and fault network construction component 12 includes: The fault event data acquisition unit is used to acquire a first fault event and determine fault intrinsic data and fault propagation data.

[0079] The fault data segmentation and feature extraction unit is used to perform multi-layer segmentation and feature extraction on the fault intrinsic data, perform hierarchical compensation based on the fault propagation data, and determine a first event segmentation tree.

[0080] The node feature filtering unit is used to perform node feature filtering on the first event segmentation tree by setting a preset correlation to determine a first valid segmentation tree.

[0081] The node failure probability threshold identification unit is used to identify the probability threshold of node failure of the first valid segmentation tree.

[0082] In some embodiments, the fault diagnosis module building component 13 includes: The fault diagnosis sample construction unit is used to organize the fault event records and construct fault diagnosis samples, wherein the fault diagnosis samples include fault discrimination samples and failure degree analysis samples.

[0083] The diagnosis decision block training unit is used to perform supervised training based on the fault discrimination samples with the fault feature network as a reference to obtain a diagnosis decision block.

[0084] The failure decision block training unit is used to perform supervised training based on the failure degree analysis samples with the fault probability network as a benchmark to obtain a failure decision block.

[0085] The diagnostic large model construction unit is used to post-connect the failure decision block to the diagnostic decision block to determine the diagnostic large model.

[0086] In some implementations, the fault diagnosis module building component 13 further includes: A lightweight loss function construction unit is used to construct a lightweight loss function, wherein the lightweight loss function represents the transmission loss from the large diagnostic model to the fault diagnosis module, and includes a logical dimension and a data dimension.

[0087] The fault diagnosis module transfer training unit is used to use the output of the large diagnosis model as the first sample and the fault diagnosis sample as the second sample, and perform transfer training in combination with the lightweight loss function to determine the fault diagnosis module.

[0088] In some implementations, the execution step of the fault diagnosis module transfer training unit in the fault diagnosis module building component 13 further includes: A first fault diagnosis sample is obtained, the diagnosis big model is input, first diagnosis data is determined and added to the first sample, wherein the first diagnosis data includes the entire diagnosis process based on the diagnosis big model.

[0089] A learning rate is set for the first fault diagnosis sample and the first diagnostic data, and a lightweight training is performed on the fault diagnosis module with the lightweight loss function as a constraint.

[0090] In some embodiments, the front-end sensor monitoring and control component 14 includes: The CNC machine tool operation mechanism acquisition unit is used to acquire the operation mechanism of the target CNC machine tool and determine multiple groups of operation elements by performing scenario decomposition.

[0091] The scene monitoring trigger identification unit is used to perform simultaneous trigger identification of scene monitoring for the multiple groups of operating elements and determine the first sampling rule.

[0092] The operating factor fluctuation characteristic analysis unit is used to traverse the multiple groups of operating factors, set the sampling frequency based on the factor fluctuation characteristics, and determine the second sampling rule, wherein the factor fluctuation is positively correlated with the sampling frequency.

[0093] The front-end sensor configuration unit is used to configure the front-end sensor based on the first sampling rule and the second sampling rule, wherein the monitoring sampling control of the front-end sensor is performed in multiple threads.

[0094] In some embodiments, the fault diagnosis execution component 15 includes: The machine tool operation data preprocessing unit is used to preprocess the machine tool operation data and transmit it to the fault diagnosis module, wherein data screening, dimensionality reduction and feature extraction are used as preprocessing standards.

[0095] The fault diagnosis decision unit is used to diagnose the occurrence of a fault and determine the diagnosis result of the fault event based on the diagnosis decision block of the fault diagnosis module.

[0096] The fault diagnosis result judgment unit is used to terminate the diagnosis process if the diagnosis result is empty.

[0097] The fault failure level determination unit is used to transfer the diagnosis result to the failure decision block if the diagnosis result is not empty, perform fault failure level determination, and output the fault diagnosis result.

[0098] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the intelligent fault diagnosis system for automated CNC machine tools described in embodiment two. For the sake of brevity of the specification, they will not be further elaborated here.

[0099] It should be understood that the embodiments disclosed in the present invention and the above description can enable those skilled in the art to use the present invention to implement the present invention. At the same time, the present invention is not limited to the above-mentioned embodiments. It should be understood that those skilled in the art can still modify the technical solutions recorded in the above-mentioned embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. An intelligent fault diagnosis method for automated CNC machine tools, characterized in that: The method comprises: Interactive target CNC machine tool fault event record; Based on the fault event records, fault characteristics mining is performed to construct a fault network, wherein the fault characteristic network is constructed by performing graph theory analysis based on fault propagation, and the fault probability network is constructed by performing fault failure analysis; According to the fault network, a large diagnostic model is trained and lightweighted to build a fault diagnosis module, wherein the module is built by two-step training based on a lightweight loss function; Determine the staggered sampling frequency, monitor and control the front-end sensor based on the staggered sampling frequency, and determine the machine tool operation data; The machine tool operation data is transmitted back to the fault diagnosis module to perform fault occurrence determination and failure level measurement to determine the fault diagnosis result.

2. The intelligent fault diagnosis method for an automated numerically controlled machine tool according to claim 1, characterized in that: Based on the fault event record, fault feature mining is performed, including: Obtaining a first fault event, and determining fault intrinsic data and fault propagation data; Performing multi-layer segmentation and feature extraction on the fault intrinsic data, performing hierarchical compensation based on the fault propagation data, and determining a first event segmentation tree; By setting a preset correlation, filtering node features of the first event segmentation tree to determine a first valid segmentation tree; A probability threshold of node failure is marked for the first valid partition tree.

3. The intelligent fault diagnosis method for an automated numerically controlled machine tool according to claim 1, characterized in that: The diagnostic large model training comprises: Arranging the fault event records and constructing fault diagnosis samples, wherein the fault diagnosis samples include fault discrimination samples and failure degree analysis samples; Taking the fault feature network as a benchmark, supervised training is performed based on the fault discrimination samples to obtain a diagnosis decision block; Taking the fault probability network as a benchmark, supervised training is performed based on the failure degree analysis samples to obtain a failure decision block; The failure decision block is post-connected to the diagnosis decision block to determine a large diagnosis model.

4. The intelligent fault diagnosis method for an automated numerically controlled machine tool as claimed in claim 3, characterized in that: Perform lightweight processing and build a fault diagnosis module, including: Constructing a lightweight loss function, wherein the lightweight loss function represents the transmission loss from the large diagnostic model to the fault diagnosis module, and includes a logical dimension and a data dimension; The output of the large diagnosis model is used as the first sample, the fault diagnosis sample is used as the second sample, and transfer training is performed in combination with the lightweight loss function to determine the fault diagnosis module.

5. The intelligent fault diagnosis method for an automated numerically controlled machine tool as claimed in claim 4, characterized in that: Combined with the lightweight loss function, transfer training is performed, including: Obtain a first fault diagnosis sample, input the diagnostic big model, determine first diagnostic data and add it into the first sample, wherein the first diagnostic data includes the entire diagnostic process based on the diagnostic big model; A learning rate is set for the first fault diagnosis sample and the first diagnostic data, and a lightweight training is performed on the fault diagnosis module with the lightweight loss function as a constraint.

6. The intelligent fault diagnosis method for an automated numerically controlled machine tool according to claim 1, characterized in that: Before determining the staggered sampling frequency, include: Obtain the operating mechanism of the target CNC machine tool, and determine multiple groups of operating elements by performing scenario decomposition; For the multiple groups of operating elements, a simultaneous trigger mark for scene monitoring is performed to determine a first sampling rule; Traversing the plurality of groups of operating elements, setting a sampling frequency based on element fluctuation characteristics, and determining a second sampling rule, wherein the element fluctuation is positively correlated with the sampling frequency; Based on the first sampling rule and the second sampling rule, the front-end sensor is configured, wherein monitoring sampling control of the front-end sensor is performed in multiple threads.

7. The intelligent fault diagnosis method for an automated numerically controlled machine tool as claimed in claim 3, characterized in that: The fault occurrence determination and failure level measurement include: Preprocessing the machine tool operation data and transmitting it to the fault diagnosis module, wherein data screening, dimensionality reduction and feature extraction are used as preprocessing criteria; Based on the diagnosis decision block of the fault diagnosis module, diagnose the occurrence of the fault and determine the diagnosis result of the fault event; If the diagnosis result is empty, terminate the diagnosis process; If the diagnosis result is not empty, the diagnosis result is transferred to the failure decision block to perform a fault failure level determination and output the fault diagnosis result.

8. Intelligent fault diagnosis system for automated CNC machine tools, characterized in that: The system is used to execute the intelligent fault diagnosis method for an automated numerically controlled machine tool according to any one of claims 1 to 7, and the system comprises: CNC machine tool fault event interaction component, used to interact with target CNC machine tool fault event records; A fault characteristic mining and fault network construction component is used to mine fault characteristics and construct a fault network based on the fault event record, wherein the fault characteristic network is constructed by performing graph theory analysis based on fault propagation, and the fault probability network is constructed by performing fault failure analysis; A fault diagnosis module building component is used to perform large-scale diagnostic model training and lightweight processing according to the fault network, and build a fault diagnosis module, wherein the module is built by two-step training based on a lightweight loss function; A front-end sensor monitoring and control component is used to determine the staggered sampling frequency, monitor and control the front-end sensor based on the staggered sampling frequency, and determine the machine tool operation data; The fault diagnosis execution component is used to transmit the machine tool operation data back to the fault diagnosis module, perform fault occurrence judgment and failure level measurement, and determine the fault diagnosis result.

Images