Electric screw press fault diagnosis method and system based on multi-feature fusion

Abstract

The invention belongs to the technical field of fault diagnosis, and particularly relates to an electric screw press fault diagnosis method and system based on multi-feature fusion, and the method comprises the steps: constructing a digital twin model of an electric screw press, generating a simulation data set through injecting different working conditions, and constructing a working condition source data set in combination with real data; a DCNN-Informer-MSAM network fault diagnosis model based on multi-feature fusion is constructed, and a working condition source data set is adopted to train the fault diagnosis data driving model; and generating a fault diagnosis result of the electric screw press based on the fault diagnosis data driving model and the currently monitored real data. And performing similarity calculation on the diagnosis result and the maintenance source data set according to a set algorithm, and screening out an optimal solution so as to realize operations such as start and stop of equipment components. The operation state of the equipment can be adjusted in real time in a complex environment, and the reliability and the operation stability of the production equipment are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and particularly relates to a fault diagnosis method and system for an electric screw press based on multi-feature fusion. Background Technique

[0002] As one of the core equipment for forging production, the electric screw press has many remarkable advantages, such as precisely controllable slider impact energy, high production efficiency, low defective rate, long die life, energy conservation and environmental protection, and low noise. These advantages have enabled it to occupy a crucial position in the die forging field of industrial mother machines. Currently, intelligent servo direct-drive numerical control die forging equipment, as a revolutionary technology, has been monopolized by foreign enterprises, seriously restricting the technological progress of our country in the field of high-precision and advanced industrial manufacturing. In the process of manufacturing and operation and maintenance of intelligent servo direct-drive numerical control electric screw presses in our country, a series of challenges such as immature fault repair and diagnosis technologies for electric screw presses still exist. Therefore, realizing the fault diagnosis and intelligent regulation of key components of the electric screw press has become the key to ensuring the continuity of the production process, improving the quality of forgings, enhancing the reliability of equipment operation, and reducing maintenance and repair costs. This not only plays a crucial role in promoting the sustainable development of our country's equipment manufacturing industry, but also has profound strategic significance.

[0003] In recent years, there are still some challenges in the research and application of fault diagnosis at home and abroad, such as insufficient consideration of fault time series information, imbalance between fault data and normal working condition data, etc. In view of these problems, it is necessary to combine the emerging technology of digital twin and propose a fault diagnosis method and system for an electric screw press based on multi-feature fusion. Summary of the Invention

[0004] The purpose of the present invention is to provide a fault diagnosis method and system for an electric screw press based on multi-feature fusion. Based on the DCNN-Informer-MSAM network, the working condition source data is weighted and fused to deeply explore the correlation between different features and potential fault information, thereby effectively solving the problem of information loss caused by a single feature dimension in traditional fault diagnosis methods, being able to adjust the operation state of the equipment in real time in a complex environment, and significantly improving the reliability and operation stability of production equipment.

[0005] To achieve the above object, the present invention provides a fault diagnosis method for an electric screw press based on multi-feature fusion, including the following steps:

[0006] S1. Construct a digital twin model of the electric screw press, generate a simulation data set by injecting different working conditions, and construct a working condition source data set in combination with the real data of the electric screw press;

[0007] S2. Construct a fault diagnosis data-driven model and train the fault diagnosis data-driven model using the working condition source dataset; the fault diagnosis data-driven model includes a DCNN-Informer-MSAM network based on multi-feature fusion, a fully connected layer, and a Softmax layer;

[0008] S3. Based on the fault diagnosis data-driven model and the currently monitored real data, obtain the frequency domain and time domain features of the real data through the DCNN and Informer networks respectively, and then input them into the MSAM multi-head task self-attention network for multi-feature fusion, thereby generating the fault diagnosis result of the electric screw press.

[0009] Further, in step S1, divide the electric screw press into four subsystems: a mechanical system, a hydraulic system, an electrical system, and a measurement and control system, and construct digital twin simulation models for different subsystems;

[0010] According to the structured data under the real working conditions of the electric screw press, adjust the subsystem parameters, and jointly use the simulation model to obtain simulation data, and compare it with the real data to obtain the fitting curve and similarity degree between the real data and the simulation data;

[0011] If the simulation curve is consistent with the actual curve, end. If there is a deviation, adjust the parameters and return to the digital twin simulation model to adjust the parameters and structure until the simulation curve is consistent with the actual curve to obtain the digital twin simulation model.

[0012] Further, in step S1, the real data includes the operating parameters of the mechanical system, hydraulic system, electrical system, and measurement and control system;

[0013] The operating parameters of the mechanical system include slider displacement, flywheel torque, and guide rail friction;

[0014] The operating parameters of the hydraulic system include hydraulic cylinder pressure, hydraulic oil temperature, and oil return flow rate;

[0015] The operating parameters of the electrical system include motor current, motor voltage, and input / output power;

[0016] The operating parameters of the measurement and control system include warning information and alarm signals;

[0017] The ratio of the amount of simulation data to real data in the working condition source dataset is 8:(1 - 3). Further, in step S2, the training process of the fault diagnosis data-driven model includes:

[0018] S21. Perform time domain and frequency domain conversions on the time series working condition data in the working condition source dataset to extract dual features in the time domain and frequency domain;

[0019] S22. Input the converted frequency-domain signal into a DCNN (Deep Convolutional Neural Network) for frequency-domain feature learning to extract the frequency-domain features of the working conditions of the electric screw press;

[0020] S23. Input the converted time-domain signal into an Informer network for time-domain feature learning. The Informer captures the long-range dependencies in the time-series data to accurately identify the changing features of the working conditions of the electric screw press in the time domain;

[0021] S24. Input the features learned separately from the time domain and the frequency domain into a MSAM (Multi-Head Task Self-Attention Network) to achieve the fusion of multi-dimensional features;

[0022] S25. Input the fused multi-dimensional features into a fully connected layer and a Softmax layer to complete the accurate classification of the equipment working conditions and fault prediction.

[0023] Further, in step S21, the frequency-domain conversion is performed by continuous wavelet transform, and the time domain is decomposed by Fourier-Moiré decomposition; the fault diagnosis data-driven model is also optimized by a multi-objective particle swarm optimization algorithm;

[0024] In step S3, input the real-time data of the currently monitored electric screw press into the trained fault diagnosis data-driven model to obtain the fault diagnosis result.

[0025] Further, the fault diagnosis method further includes the following steps:

[0026] S4. Combine the maintenance work orders and working characteristics of the electric screw press to generate a maintenance source dataset;

[0027] S5. Calculate the similarity between the fault diagnosis result of step S3 and the maintenance source dataset according to a set algorithm, and screen out the optimal solution;

[0028] S6. According to the screened solution, send an operation instruction to the control device to realize the start-stop operation of each subsystem of the electric screw press.

[0029] Further, in step S3, the construction process of the maintenance source dataset includes the following steps:

[0030] S41. Combine historical maintenance event work orders and working characteristics to classify the fault categories, and screen out the valid event work orders containing actual solutions;

[0031] S42. Extract the fault characteristics of the electric screw press and their corresponding maintenance solutions from the screened valid event work orders to form key knowledge points;

[0032] S43. Organize the extracted fault characteristics and maintenance plan business knowledge points according to preset rules to generate a complete knowledge point set, and build a professional knowledge base for the fault diagnosis of electric screw presses based on this knowledge point set;

[0033] The preset rules build a structured data set by splitting the maintenance work order into three parts: "fault problem - solution measure - working condition data", screen out data pairs in the form of fault problem - solution measure - working condition data that are effective from this data set, and summarize and organize them to generate a standardized knowledge point set.

[0034] Further, in step S5, the setting algorithm includes the following steps:

[0035] S51. Calculate the similarity between the fault diagnosis result and the maintenance source data set, analyze and obtain the similarity score of the equipment fault, and screen out the knowledge points that best match the current fault according to the score;

[0036] The calculation process of the similarity includes: converting the fault situation of the electric screw press and the maintenance source data set into TF-IDF vectors respectively, and calculating the similarity between the fault situation of the electric screw press and each knowledge point in the maintenance source data set through cosine similarity;

[0037] S52. Concatenate the screened matching fault knowledge points with the diagnosis result to generate a Prompt text;

[0038] S53. Input the generated Prompt text into the large language model, and through the processing and analysis of the model, output the diagnosis result and the fault solution.

[0039] The present invention also provides a fault diagnosis system for electric screw presses based on multi-feature fusion, including:

[0040] An intelligent data monitoring module, which is used to monitor the operation data of the electric screw press in real time, generate a simulation data set according to the digital twin model of the electric screw press, and further build a working condition source data set;

[0041] A fault diagnosis module, which is used to perform fault diagnosis on the electric screw press according to the fault diagnosis data-driven model of the DCNN-Informer-MSAM network based on multi-feature fusion and the monitored operation data of the electric screw press to obtain a diagnosis result; the fault diagnosis data-driven model includes a DCNN-Informer-MSAM network based on multi-feature fusion, a fully connected layer, and a Softmax layer.

[0042] Furthermore, the fault diagnosis system further includes an intelligent control strategy module, which is used to combine the working condition source data set and the maintenance source data set, calculate the similarity between the fault diagnosis result and the fault knowledge point set in the maintenance source data through a set algorithm, screen out the optimal solution, and generate control instructions and warning information;

[0043] The equipment control execution module is used to receive the solution and control instructions provided by the intelligent control strategy module, communicate with each control device through the data communication interface, and realize start, stop, and adjustment operations to ensure that each control device operates according to the preset strategy.

[0044] Furthermore, the intelligent data monitoring module includes: a data acquisition unit, a data positioning unit, and a data processing unit; the data acquisition unit is used to acquire the operation data of the mechanical system, hydraulic system, electrical system, and measurement and control system; the data positioning unit is used to configure and manage the specific installation position information of different sensors on the electric screw press to ensure the accuracy and pertinence of data acquisition;

[0045] The intelligent control strategy module includes: a data receiving unit, a characteristic matching unit, and an instruction generating unit;

[0046] The equipment control execution module includes: an instruction parsing unit, a communication interface unit, a safety protection unit, and a status monitoring unit;

[0047] Among them, the instruction generating unit includes:

[0048] The area identification and instruction allocation unit is used to refine the identification of the equipment fault area, automatically adjust the switch state and operation parameters of the control devices in each area according to the area division and real-time working condition monitoring data, and ensure the efficient operation of the equipment;

[0049] The fault monitoring and instruction regulation unit receives the data from the intelligent data monitoring module. When a control device in a certain area fails, based on the real-time working condition data provided by the data processing unit and the equipment characteristic information provided by the characteristic matching unit, it automatically adjusts the actions of the relevant control devices to reduce the impact of the fault and maintain the stable operation of the equipment.

[0050] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following technical advantages are mainly possessed:

[0051] 1. The fault diagnosis method of the electric screw press based on multi - feature fusion provided by the present invention is based on the DCNN - Informer - MSAM fault diagnosis network for multi - feature fusion. This network is based on the constructed working condition source data set, comprehensively considering various signal characteristics of the electric screw press. In this framework, DCNN extracts features from the signal transformed to the frequency domain through continuous wavelet transform (CWT); Informer extracts features from the time - domain signal obtained by Fourier - Moiré decomposition (FMD), and then further performs weighted fusion on the extracted features; MSAM accurately captures key information in the data and performs effective weighting through a multi - scale attention mechanism. This method can deeply explore the correlation between different features and potential fault information, thus effectively solving the problem of information loss caused by a single feature dimension in traditional fault diagnosis methods. At the same time, the network can better learn and highlight important fault characteristics, significantly improving the interpretability and diagnostic accuracy of the fault diagnosis network.

[0052] 2. The present invention divides the electric screw press into four subsystems: mechanical system, hydraulic system, electrical system, and measurement and control system, which facilitates data acquisition, classification, and management, enabling more accurate and efficient identification of the equipment operation state, and thus improving the fault diagnosis efficiency and accuracy.

[0053] 3. The present invention also constructs a maintenance source data set, fuses equipment maintenance event work orders and equipment working characteristics, and generates a complete fault diagnosis model and professional knowledge base based on this. The constructed fault diagnosis system can monitor the equipment state in real - time and effectively, significantly improving the accuracy and automation level of fault diagnosis, providing strong support for equipment monitoring and maintenance.

[0054] 4. The fault diagnosis system based on multi - feature fusion provided by the present invention combines the multi - feature fusion fault diagnosis algorithm with the LLaMa2 large language model to realize the self - diagnosis and intelligent control of the electric screw press. This diagnosis system can monitor the equipment operation state in real - time and effectively, significantly improving the accuracy and automation level of fault diagnosis, providing solid technical support for the intelligent monitoring and maintenance of equipment, and promoting the progress of the intelligent manufacturing field. Brief Description of the Drawings

[0055] Figure 1 It is a flow chart of the fault diagnosis method of the electric screw press based on multi - feature fusion in Embodiment 1 of the present invention.

[0056] Figure 2 It is a structural diagram of the fault diagnosis system of the electric screw press based on multi - feature fusion in Embodiment 2 of the present invention.

[0057] Figure 3This is the flowchart of the self-control method for an electric screw press based on multi-feature fusion in Embodiment 3 of the present invention.

[0058] Figure 4 This is the flowchart of the DCNN-Informer-MSAM network diagnosis method based on multi-feature fusion of the present invention. Detailed implementation manners

[0059] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0060] Embodiment 1

[0061] As Figure 1 shown, an electric screw press fault diagnosis method based on multi-feature fusion provided by an embodiment of the present invention includes:

[0062] S1. Construct a digital twin model of the electric screw press, generate a simulation data set by injecting different working conditions, and construct a working condition source data set in combination with the real data of the electric screw press;

[0063] S2. Construct a fault diagnosis data-driven model, and train the fault diagnosis data-driven model with the working condition source data set; the fault diagnosis data-driven model includes a DCNN-Informer-MSAM network based on multi-feature fusion, a fully connected layer, and a Softmax layer;

[0064] S3. Based on the fault diagnosis data-driven model and the currently monitored real data, obtain the frequency domain and time domain features of the real data through the DCNN and Informer networks (time series prediction models) respectively, and then input them into the MSAM (Multi-Scale Channel Attention Module) multi-head task self-attention network for multi-feature fusion, and then generate the fault diagnosis result of the electric screw press;

[0065] S4. Combine the maintenance work orders and working characteristics of the electric screw press to generate a maintenance source data set;

[0066] S5. Calculate the similarity between the fault diagnosis result in step S3 and the maintenance source data set according to the set algorithm, and screen out the optimal solution;

[0067] S6. According to the selected solution, send an operation instruction to the control device to realize the start and stop operations of each subsystem of the electric screw press.

[0068] Specifically, in step S1, the press subsystem is divided into mechanical system, hydraulic system, electrical system and measurement and control system, and digital twin simulation models of different subsystems are constructed to ensure the high fidelity of the digital twin model. According to the structured data of the electric screw press under the actual working conditions (i.e., the operating parameters of each subsystem), the subsystem parameters are adjusted, and the simulation data is obtained by combining the simulation model, and compared with the real data to obtain the fitting curve and similarity between the real data and the simulation data;

[0069] If the simulation curve is consistent with the actual curve, the process ends; if there is a deviation, the parameters are adjusted and the digital twin simulation model is returned to adjust the parameters and structure until the simulation curve is consistent with the actual curve to obtain a digital twin simulation model.

[0070] The mechanical system is mainly used to describe the composition of a mechanical structure. This system mainly uses mechanical transmission principles and mechanics principles to express and complete certain tasks. The hydraulic system is used to describe a motion control system that uses oil as a medium to transfer energy.

[0071] Specifically, the real data includes the operating parameters of the mechanical system, hydraulic system, electrical system, and measurement and control system;

[0072] The operating parameters of the mechanical system include slider displacement, flywheel torque, and guide rail friction;

[0073] The operating parameters of the hydraulic system include hydraulic cylinder pressure, hydraulic oil temperature, and return oil flow rate;

[0074] The operating parameters of the electrical system include motor current, motor voltage, input and output power;

[0075] The operating parameters of the measurement and control system include early warning information and alarm signals;

[0076] The ratio of the amount of simulation data to the amount of real data in the operating condition source data set is 8:(1-3), preferably 8:2.

[0077] Specifically, the DCNN-Informer-MSAM network based on multi-feature fusion includes a frequency domain feature extractor based on DCNN, a time domain feature extractor based on Informer, and a multi-dimensional feature fuser based on MSAM network.

[0078] See also Figure 4 In step S2, the training process of the fault diagnosis data driven model includes:

[0079] S21, converting the time series operating condition data in the operating condition source data set into time domain and frequency domain to extract dual features in time domain and frequency domain;

[0080] S22. Input the converted frequency-domain signal into the DCNN (Deep Convolutional Neural Network) for frequency-domain feature learning to extract the frequency-domain features of the electric screw press operating conditions. The deep convolutional neural network learns local features through convolutional layers and reduces the dimension through pooling layers to extract local key features.

[0081] S23. Input the converted time-domain signal into the Informer network for time-domain feature learning. The Informer captures the long-range dependence relationships in the time-series data to accurately identify the change features of the electric screw press operating conditions in the time domain. The Informer adopts a self-attention mechanism combined with the ProbSparse mechanism and convolutional feature extraction to enhance the modeling ability for long-term dependence relationships and the time-series feature extraction ability.

[0082] S24. Input the features learned separately from the time domain and the frequency domain into the MSAM (Multi-Head Task Self-Attention Network) to achieve the fusion of multi-dimensional features.

[0083] S25. Input the fused multi-dimensional features into the fully connected layer and the Softmax layer (normalized exponential function) to complete the accurate classification of the equipment operating conditions and fault prediction.

[0084] Furthermore, in step S21, the frequency-domain conversion is performed through continuous wavelet transform (CWT), and the time domain is decomposed through Fourier-Moiré decomposition (FMD decomposition). In step S24, a cross-modal attention mechanism is adopted for multi-dimensional feature fusion. This module fuses the frequency-domain features extracted by the DCNN and the time-domain features extracted by the Informer through a weighting strategy, screens out the most important features from them, and passes them to the subsequent network layers.

[0085] The fault diagnosis data-driven model is also optimized through a multi-objective particle swarm optimization algorithm.

[0086] Specifically, the multi-objective particle swarm optimization specifically includes the following steps:

[0087] S251: Initialize the velocities and positions of all particles, determine the constraint conditions, initialize the external archive, and form the Pareto solution set. The particles include the number and size of the convolutional layers in the DCNN, the length of the time-series modeling in the Informer model, the learning rate, and the parameters of each layer's attention mechanism in the MSAM. The constraint conditions include accuracy constraint, recall constraint, F1-score constraint, convergence constraint, and hyperparameter constraint.

[0088] S252: Calculate the objective function values of all particles, screen out the non-dominated solutions according to the Pareto dominance relationship, and store them in the external archive.

[0089] S253: Select non-dominated solutions from the external archive as candidate global optimal solutions, and determine the global optimal particle based on the crowding degree selection strategy;

[0090] S254: Update the velocity and position of the particle. If the particle exceeds the search space boundary, strategies such as reflection boundary, random reset, or velocity zeroing are used for processing;

[0091] S255: Update the individual optimal position of each particle according to the Pareto domination principle;

[0092] S256: If the maximum iteration number T is reached or the solutions in the external archive have converged, terminate the algorithm; otherwise, return to step S252 to continue the iteration.

[0093] In step S3, input the real-time data of the electric screw press currently monitored into the trained fault diagnosis data-driven model to obtain the fault diagnosis result. Specifically:

[0094] S31: Perform time-domain and frequency-domain conversions on the time-series working condition data of the electric screw press monitored in real time to extract dual features in the time domain and frequency domain; thereby better capturing potential patterns in the signal and providing richer information for subsequent feature learning.

[0095] S32: Input the converted frequency-domain signal into the DCNN deep convolutional neural network for frequency-domain feature learning to extract the frequency-domain features of the working conditions of the electric screw press;

[0096] S33: Input the converted time-domain signal into the Informer network for time-domain feature learning. Through the Informer, capture the long-range dependence relationships in the time-series data to accurately identify the change features of the working conditions of the electric screw press in the time domain;

[0097] S34: Input the features learned from the time domain and frequency domain respectively into the MSAM multi-head task self-attention network to achieve the fusion of multi-dimensional features; improve the comprehensive ability of feature representation, enhance the discriminant performance of the model, and thus improve the accuracy and robustness of fault diagnosis.

[0098] S35: Input the fused multi-dimensional features into the fully connected layer and the Softmax layer to complete the accurate classification of the equipment working conditions and fault prediction, providing a reliable basis for fault diagnosis.

[0099] Specifically, the DCNN-Informer-MSAM network architecture is as follows:

[0100] Table 1 DCNN architecture

[0101] Layer type Input dimension Output dimension Convolutional layer 1 (32,20,32) (30,18,32) Pooling layer 1 (30,18,32) (15,9,32) Convolutional layer 2 (15,9,32) (13,7,64) Pooling layer 2 (13,7,64) (6,3,64) Convolutional layer 3 (6,3,64) (4,1,128) Linear layer (4,1,128) 128

[0102] Table 2 Informer Architecture

[0103] Input dimension (32,20,32) Output dimension 128 Number of hidden layers 4

[0104] Table 3 MSAM Architecture

[0105] Input dimension 128 Output dimension 8

[0106] Specifically, in the step S4, the construction process of the maintenance source dataset includes the following steps:

[0107] S41: Combine historical maintenance event work orders and equipment working characteristics to classify equipment failure categories, and screen out valid event work orders containing actual solutions;

[0108] S42: Extract the failure characteristics of the electric screw press and their corresponding maintenance plans from the screened event work orders to form key knowledge points;

[0109] S43: Organize the extracted failure characteristics and maintenance plan business knowledge points according to preset rules to generate a complete knowledge point set; and build a professional knowledge base for electric screw press fault diagnosis based on this knowledge point set.

[0110] Specifically, in the step S43, the preset rules are to split the maintenance work order into three parts: "fault problem - solution measure - working condition data" to construct a structured dataset. Screen out valid "fault problem - solution measure - working condition data" data pairs from this dataset, and summarize and organize them to generate a standardized knowledge point set.

[0111] Specifically, in the step S5, the set algorithm includes the following steps:

[0112] S51: Calculate the similarity between the fault diagnosis result and the maintenance source dataset, analyze and obtain the similarity score of the equipment fault, and screen out the knowledge points most matching the current fault according to the score;

[0113] S52: Concatenate the screened matching fault knowledge points with the diagnosis result to generate a Prompt text;

[0114] S53: Input the generated Prompt text into the large language model, and through the processing and analysis of the model, output the diagnosis result and the fault solution.

[0115] Specifically, in the step S51, the similarity calculation process includes: converting the fault situation of the electric screw press and the maintenance source dataset into TF-IDF vectors respectively, and calculating the similarity between the fault situation of the electric screw press and each knowledge point in the maintenance source dataset through cosine similarity.

[0116] The TF-IDF vector conversion formula is as follows:

[0117] TF-IDF = TF · IDF

[0118] Where:

[0119]

[0120] In the formula, n i,j represents the number of times a term t i appears in a document d j , and TF i,j represents the frequency of the term t i appearing in the document d j .

[0121]

[0122] In the formula, |D| represents the number of all documents; |j:t i ∈d j | represents the number of documents containing the term t i ; IDF i represents the uniqueness of the keyword i in the entire document set.

[0123] The cosine similarity calculation formula is as follows:

[0124]

[0125] In the formula, Similarity is the cosine similarity between the maintenance source data set and the diagnosis result; A is the converted maintenance source data set; B is the converted diagnosis result; ∥A∥ is the norm of the converted maintenance source data set; ∥B∥ is the norm of the converted diagnosis result.

[0126] Example 2

[0127] As Figure 2 shown, the embodiment of the present invention further provides a fault diagnosis system for an electric screw press based on multi-feature fusion, including:

[0128] An intelligent data monitoring module, which is used to monitor the operation data of the electric screw press in real time, generate a simulation data set according to the digital twin model of the electric screw press, and further construct a working condition source data set;

[0129] The fault diagnosis module is used to perform fault diagnosis on the electric screw press according to the fault diagnosis data-driven model of the DCNN-Informer-MSAM network based on multi-feature fusion and the monitored operation data of the electric screw press, and obtain a diagnosis result; the fault diagnosis data-driven model includes a DCNN-Informer-MSAM network based on multi-feature fusion, a fully connected layer, and a Softmax layer.

[0130] Preferably, it further includes an intelligent control strategy module, which is used to combine the working condition source data set and the maintenance source data set, calculate the similarity between the fault diagnosis result and the fault knowledge point set in the maintenance source data set through a set algorithm, screen out the optimal solution, and generate control instructions and warning information;

[0131] The equipment control execution module is used to receive the solution and control instructions provided by the intelligent control strategy module, and communicate with each control device through a data communication interface to implement start, stop, and adjustment operations, ensuring that each control device operates according to the preset strategy.

[0132] Specifically, the intelligent data monitoring module includes: a data acquisition unit, a data positioning unit, and a data processing unit; the data acquisition unit is used to acquire the operation data of the mechanical system, hydraulic system, electrical system, and measurement and control system; the data positioning unit is used to configure and manage the specific installation position information of different sensors on the electric screw press to ensure the accuracy and pertinence of data acquisition; the data processing unit is used to process the data to obtain a working condition source data set. In particular, a maintenance source data set is also obtained according to the historical maintenance work orders and working characteristics of the electric screw press.

[0133] The intelligent control strategy module includes: a data receiving unit, a characteristic matching unit, and an instruction generating unit; the equipment control execution module includes: an instruction parsing unit, a communication interface unit, a safety protection unit, and a status monitoring unit.

[0134] Specifically, the instruction generating unit includes:

[0135] The area identification and instruction allocation unit: refine the identification of the equipment fault area, and automatically adjust the switch states and operation parameters of the control devices in each area according to the area division and real-time working condition monitoring data to ensure the efficient operation of the equipment.

[0136] The fault monitoring and instruction regulation unit: receive data from the equipment status monitoring module. When a control device in a certain area fails, based on the real-time working condition data provided by the data processing unit and the equipment characteristic information provided by the characteristic matching unit, automatically adjust the actions of the relevant control devices to reduce the impact of the fault and maintain the stable operation of the equipment.

[0137] Embodiment 3

[0138] like Figure 3 As shown, the embodiment of the present invention also provides an electric screw press automatic control method based on multi-feature fusion, comprising:

[0139] S301: monitor the operating status of the electric screw press equipment, collect various operating parameters, and transmit the data to the control system for real-time analysis and processing;

[0140] S302: Based on the DCNN-Informer-MSAM network with multi-feature fusion, potential fault risks and performance deviations are identified, and fault information is generated by combining historical data and preset process requirements;

[0141] S303: Using a large language model to intelligently evaluate equipment fault information and operating status, automatically generate warning information and propose optimization strategies (fault solutions);

[0142] S304: According to the optimization strategy, an operation instruction is issued to the control system to execute the start and stop operation of equipment components, speed regulation and other key parameter adjustments, correct process deviations or trigger fault handling procedures to ensure the stability and reliability of equipment operation;

[0143] S305: Through continuous learning and optimization of control strategies, the large language model combines the historical operation data and real-time feedback of the equipment to continuously update and adjust the control algorithm to improve the intelligence level and adaptability of the automatic control system;

[0144] S306: The system generates a detailed operation report for operators to refer to, to assist them in equipment maintenance, optimize production processes or promote subsequent equipment improvements.

[0145] Specifically, the optimization control strategy in step S305 includes control optimization based on physical models, control optimization based on data-driven, adaptive control strategy optimization, self-regulating control, Pareto optimization and other methods.

[0146] Specifically, the content of the operation report in step S306 includes equipment health status, fault warning, adjustment measures, historical trends and the like.

[0147] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A fault diagnosis method for electric screw press based on multi-feature fusion, characterized in that: The following steps are involved: S1. Build a digital twin model of the electric screw press, generate a simulation data set by injecting different working conditions, and build a working condition source data set in combination with the real data of the electric screw press; S2. Constructing a fault diagnosis data-driven model, and using the working condition source data set to train the fault diagnosis data-driven model; the fault diagnosis data-driven model includes a DCNN-Informer-MSAM network based on multi-feature fusion, a fully connected layer and a Softmax layer; S3. Based on the fault diagnosis data-driven model and the real data currently monitored, the frequency domain and time domain features of the real data are obtained through DCNN and Informer networks respectively, and then input into the MSAM multi-task self-attention network for multi-feature fusion, thereby generating the fault diagnosis results of the electric screw press.

2. The electric screw press fault diagnosis method based on multi-feature fusion according to claim 1 is characterized in that: In step S1, the electric screw press is divided into four subsystems: a mechanical system, a hydraulic system, an electrical system, and a measurement and control system, and digital twin simulation models of different subsystems are constructed; According to the structured data of the electric screw press under the actual working conditions, the subsystem parameters are adjusted, and the simulation model is combined to obtain the simulation data, which is compared with the real data to obtain the fitting curve and similarity between the real data and the simulation data; If the simulation curve is consistent with the actual curve, the process ends; if there is a deviation, the parameters are adjusted and the digital twin simulation model is returned to adjust the parameters and structure until the simulation curve is consistent with the actual curve to obtain a digital twin simulation model.

3. The electric screw press fault diagnosis method based on multi-feature fusion according to claim 1 is characterized in that: In step S1, the real data includes operating parameters of the mechanical system, hydraulic system, electrical system and measurement and control system; The operating parameters of the mechanical system include slider displacement, flywheel torque, and guide rail friction; The operating parameters of the hydraulic system include hydraulic cylinder pressure, hydraulic oil temperature, and return oil flow rate; The operating parameters of the electrical system include motor current, motor voltage, input and output power; The operating parameters of the measurement and control system include early warning information and alarm signals; The ratio of the amount of simulated data to the amount of real data in the working condition source data set is 8:(1-3).

4. The electric screw press fault diagnosis method based on multi-feature fusion according to claim 1 is characterized in that: In step S2, the training process of the fault diagnosis data driven model includes: S21, converting the time series operating condition data in the operating condition source data set into time domain and frequency domain to extract dual features in time domain and frequency domain; S22, inputting the converted frequency domain signal into the DCNN deep convolutional neural network to perform frequency domain feature learning to extract the frequency domain features of the working condition of the electric screw press; S23, inputting the converted time domain signal into the Informer network to perform time domain feature learning, and capturing the long-range dependency in the time series data through the Informer to accurately identify the change characteristics of the working condition of the electric screw press in the time domain; S24, input the features learned from the time domain and frequency domain into the MSAM multi-task self-attention network to achieve the fusion of multi-dimensional features; S25. Input the fused multi-dimensional features into the fully connected layer and the Softmax layer to complete the accurate classification of equipment working conditions and fault prediction.

5. The electric screw press fault diagnosis method based on multi-feature fusion according to claim 4 is characterized in that: In step S21, the frequency domain conversion is performed by continuous wavelet transform, and the time domain is decomposed by Fourier-Moore decomposition; the fault diagnosis data driven model is also optimized by a multi-objective particle swarm optimization algorithm; In step S3, the real data of the electric screw press currently monitored is input into the trained fault diagnosis data-driven model to obtain a fault diagnosis result.

6. The electric screw press fault diagnosis method based on multi-feature fusion according to any one of claims 1 to 5, characterized in that: The step S3 further includes the following steps: S4, combining the maintenance work order and work characteristics of the electric screw press to generate a maintenance source data set; S5, calculating the similarity between the fault diagnosis result of step S3 and the maintenance source data set according to the set algorithm, and screening out the best solution; S6. Based on the selected solutions, an operation instruction is issued to the control device to realize the start and stop operation of each subsystem of the electric screw press.

7. The electric screw press fault diagnosis method based on multi-feature fusion according to claim 6 is characterized in that: In step S3, the process of constructing the maintenance source data set includes the following steps: S41. Combine historical maintenance event work orders and work characteristics to classify fault categories and select valid event work orders containing practical solutions; S42, extracting the fault characteristics of the electric screw press and its corresponding maintenance plan from the screened valid event work orders to form key knowledge points; S43, sorting out the extracted fault characteristics and maintenance solution business knowledge points according to preset rules to generate a complete knowledge point set, and building an electric screw press fault diagnosis professional knowledge base based on the knowledge point set; The preset rules construct a structured data set by splitting the maintenance work order into three parts: "fault problem-solution-operating condition data", and filter out valid data pairs in the form of fault problem-solution-operating condition data from the data set, summarize and organize them, and generate a standardized knowledge point set.

8. The electric screw press fault diagnosis method based on multi-feature fusion according to claim 5 is characterized in that: In step S5, the setting algorithm includes the following steps: S51, performing similarity calculation on the fault diagnosis result and the maintenance source data set, analyzing and obtaining the similarity score of the equipment fault, and selecting the knowledge point that best matches the current fault according to the score; The similarity calculation process includes: converting the fault condition and maintenance source data sets of the electric screw press into TF-IDF vectors respectively, and calculating the similarity between each knowledge point in the fault condition of the electric screw press and the maintenance source data set by cosine similarity; S52, combining the selected matching fault knowledge points with the diagnosis results to generate a prompt text; S53, input the generated Prompt text into the large language model, and output the diagnosis result and fault solution through model processing and analysis.

9. An electric screw press fault diagnosis system based on multi-feature fusion, characterized in that: include: Intelligent data monitoring module, which is used to monitor the operating data of the electric screw press in real time, generate a simulation data set based on the digital twin model of the electric screw press, and then construct a working condition source data set; The fault diagnosis module is used to diagnose the fault of the electric screw press according to the fault diagnosis data-driven model and the real data of the electric screw press monitored to obtain the diagnosis result; the fault diagnosis data-driven model includes a DCNN-Informer-MSAM network based on multi-feature fusion, a fully connected layer and a Softmax layer.

10. The electric screw press fault diagnosis system based on multi-feature fusion according to claim 9 is characterized in that: It also includes an intelligent control strategy module, which is used to combine the working condition source data set and the maintenance source data set, calculate the similarity between the fault diagnosis result and the fault knowledge point set in the maintenance source data set through a set algorithm, screen out the optimal solution, and generate control instructions and warning information; The equipment control execution module is used to receive the solutions and control instructions provided by the intelligent control strategy module, and communicate with each control device through the data communication interface to realize the start, stop and adjustment operations to ensure that each control device operates according to the preset strategy; Preferably, the intelligent data monitoring module includes: a data acquisition unit, a data positioning unit and a data processing unit; the data acquisition unit is used to collect the operation data of the mechanical system, the hydraulic system, the electrical system and the measurement and control system; the data positioning unit is used to configure and manage the specific installation position information of different sensors on the electric screw press to ensure the accuracy and pertinence of data acquisition; The intelligent control strategy module includes: a data receiving unit, a characteristic matching unit and an instruction generating unit; The equipment control execution module includes: an instruction parsing unit, a communication interface unit, a safety protection unit and a status monitoring unit; Wherein, the instruction generation unit includes: The area identification and command distribution unit is used to identify the equipment fault area in detail, and automatically adjust the switch status and operating parameters of the control equipment in each area according to the area division and real-time working condition monitoring data to ensure the efficient operation of the equipment; The fault monitoring and command control unit receives data from the intelligent data monitoring module. When a control device in a certain area fails, it automatically adjusts the action of the relevant control equipment based on the real-time operating condition data provided by the data processing unit and the equipment characteristic information provided by the characteristic matching unit to reduce the impact of the fault and maintain stable operation of the equipment.

Images