A predictive maintenance system and method for intelligent manufacturing equipment

Abstract

The application discloses a kind of predictive maintenance systems and methods of intelligent manufacturing equipment, it is related to intelligent manufacturing equipment maintenance technical field, the application includes: multi-source data processing and feature construction module, intelligent diagnosis and early warning module, decision generation module and local database, complete multi-source heterogeneous data standardization acquisition and space-time working condition processing, extract multidimensional feature after cleaning, generate equipment state feature index sequence, output fault occurrence probability and hierarchical early warning by pre-trained multi-fault mode identification model, form structured instruction with visual marker, call digital twin model to carry out fault evolution simulation and maintenance simulation, dynamically allocate tasks in combination with production plan and resource inventory, automatically generate executable maintenance scheme, effectively improve early fault identification accuracy and decision efficiency, reduce unplanned downtime, reduce maintenance resource waste, realize the upgrade of equipment operation and maintenance from experience dominant to data-driven.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing equipment maintenance technology, and specifically to a predictive maintenance system and method for intelligent manufacturing equipment. Background Technology

[0002] With the continuous evolution of intelligent manufacturing equipment operation and maintenance technology, equipment maintenance mode has gradually shifted from passive post-event maintenance and regular preventive maintenance to proactive predictive maintenance. Iteration of sensing technology has enabled the comprehensive collection of multi-source operating data of equipment, the development of data processing technology has supported the accurate extraction of multi-dimensional feature indicators, and the maturity of artificial intelligence algorithms has promoted the construction and application of multi-fault mode recognition models. Combined with the fault evolution simulation capability of digital twin technology, a predictive maintenance technology system based on data-driven, accurate prediction, and intelligent decision-making has been formed, realizing the transformation of equipment maintenance from experience-driven to technology-driven.

[0003] Existing technologies, such as the invention patent applications related to predictive maintenance of intelligent manufacturing equipment disclosed in announcement numbers CN120806914A and CN113657693B, have significant shortcomings upon comparison: most solutions lack effective spatiotemporal alignment and condition segmentation processing of multi-source heterogeneous data from manufacturing equipment sensors, actuators, and process control systems, extracting only single-dimensional features, making it difficult to accurately capture early fault signals under different process loads, resulting in large fault identification biases. At the same time, some early warning models only output general fault levels, failing to generate structured and visualized early warning instructions based on fault types, making it impossible to directly translate early warning information into executable maintenance tasks. Furthermore, existing solutions generally lack fault evolution simulation and maintenance process simulation supported by digital twin models, and maintenance decisions rely heavily on human experience, making it difficult to balance maintenance resources, production plans, and fault development trends, often resulting in unreasonable maintenance timing, which neither effectively avoids equipment downtime risks nor easily leads to resource waste. Summary of the Invention

[0004] In view of the above-mentioned technical deficiencies, the purpose of this invention is to provide a predictive maintenance system and method for intelligent manufacturing equipment.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The first aspect of the present invention provides a predictive maintenance system for intelligent manufacturing equipment, including a multi-source data processing and feature construction module: used to acquire sensor data, driver operation data and process control system data of the target manufacturing equipment through a standardized interface, and perform data preprocessing and feature extraction to generate a feature index sequence reflecting the status of the target manufacturing equipment.

[0006] Intelligent diagnosis and early warning module: used to input the feature index sequence into a pre-trained multi-fault mode recognition model, calculate the potential failure probability and early warning level of the target manufacturing equipment, and generate early warning instructions.

[0007] Maintenance Decision Generation Module: This module is used to call upon the digital twin model corresponding to the target manufacturing equipment to simulate fault evolution and maintenance process, and output predictive maintenance solutions for the target manufacturing equipment.

[0008] A second aspect of the present invention provides a method for a predictive maintenance system for intelligent manufacturing equipment, comprising step 1. Multi-source data processing and feature construction: acquiring sensor data, driver operation data and process control system data of the target manufacturing equipment through a standardized interface, and performing data preprocessing and feature extraction to generate a sequence of feature indicators reflecting the status of the target manufacturing equipment.

[0009] Step 2. Intelligent diagnosis and early warning: Input the feature index sequence into the pre-trained multi-fault mode recognition model, calculate the potential fault probability and early warning level of the target manufacturing equipment, and generate early warning instructions.

[0010] Step 3. Maintenance Decision Generation: Call the digital twin model corresponding to the target manufacturing equipment to simulate the fault evolution and maintenance process, and output a predictive maintenance plan for the target manufacturing equipment.

[0011] The beneficial effects of the present invention are as follows: (1) The first part of the present invention: breaks through the limitation of a single data dimension, integrates multiple types of operating data through a standardized interface, combines wavelet denoising and working condition segmentation technology, realizes the spatiotemporal precise alignment of multi-source heterogeneous data, and the extracted multi-dimensional features can accurately match the equipment status under different process loads, effectively capture the weak signals of early faults, greatly improve the data quality, provide a comprehensive and reliable analysis basis for subsequent fault diagnosis, and solve the problem of disconnection between traditional data processing and working conditions.

[0012] (2) The second part of the present invention: Based on the pre-trained multi-fault mode recognition model, it can output accurate fault probability and graded warning. Combined with the structured instructions with visual marking, the fault type and risk level are clear at a glance. Compared with the traditional general warning method, this module can realize accurate fault location and graded response, avoid invalid warning interference, help maintenance personnel quickly focus on core issues, and improve the timeliness and pertinence of fault response.

[0013] (3) The third part of the present invention: the introduction of digital twin model to carry out fault evolution simulation and maintenance simulation can accurately predict the fault development trend and automatically generate the optimal maintenance plan in combination with production plan and resource inventory. This method abandons the decision-making mode that relies on human experience, rationally plans maintenance priority and time window, effectively avoids the risk of unplanned equipment downtime, reduces maintenance resource waste, and achieves efficient collaboration between production and maintenance. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the system modules of the present invention.

[0016] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1 As shown, the present invention provides a predictive maintenance system for intelligent manufacturing equipment, including a multi-source data processing and feature construction module, an intelligent diagnosis and early warning module, a decision generation module, and a local database.

[0019] It should be noted that the multi-source data processing and feature construction module is connected to the intelligent diagnosis and early warning module, the intelligent diagnosis and early warning module is connected to the decision generation module, and the local database is connected to the multi-source data processing and feature construction module, the intelligent diagnosis and early warning module, and the decision generation module.

[0020] The multi-source data processing and feature construction module is used to acquire sensor data, driver operation data and process control system data of the target manufacturing equipment through a standardized interface, and to perform data preprocessing and feature extraction to generate a sequence of feature indicators reflecting the status of the target manufacturing equipment.

[0021] In a specific embodiment of the present invention, the data preprocessing and feature extraction are performed as follows: wavelet denoising, bandpass filtering and outlier removal are performed on the acquired sensor data and driver operating data; time-domain statistical features, frequency-domain spectral features and time-frequency features are extracted from the preprocessed sensor data; current harmonic features and sideband features characterizing the mechanical transmission state are extracted from the preprocessed driver operating data by demodulating and analyzing the current signal; torque pulsation features are extracted from the speed and torque signals; and instantaneous efficiency features are extracted based on electrical parameters.

[0022] The process control system data is synchronized and aligned based on the master clock timestamp, and the continuous operation data is divided into operating stages with different process loads according to the production formula parameters and equipment start-stop signals. Statistical features are extracted from the preprocessed and segmented process control system data to characterize the load intensity and working mode of the target manufacturing equipment under each operating stage.

[0023] The production formula parameters refer to a set of core process settings preset in the equipment process control system to execute specific production procedures. They directly determine the load and mode of equipment operation and are key to linking sensor data with production operations and realizing working condition segmentation. The specific content varies from equipment to equipment, such as the spindle speed and feed rate of CNC machine tools, and the mold temperature and injection pressure of injection molding machines. By monitoring the switching of these parameters, different process stages can be automatically identified, providing a status basis for subsequent feature extraction and diagnosis.

[0024] It should be noted that wavelet denoising technology is used to suppress various random noises and electromagnetic interferences, while retaining the true impulse and periodic components in the signal. By setting a bandpass filter with an appropriate cutoff frequency, high-frequency noise and low-frequency drift unrelated to equipment operation and fault characteristics are filtered out. Furthermore, outlier removal is performed on the collected data based on statistical methods, such as the 3σ criterion, to eliminate pseudo-data caused by sensor transient failures or communication errors. These steps are routine methods for industrial data cleaning by those skilled in the art. The specific parameters, such as wavelet basis functions and filter bands, can be adapted to the specific model and operating frequency band of the target equipment. The core purpose of preprocessing is to provide a clean and reliable input signal for subsequent feature extraction.

[0025] The time-domain statistical features extracted from the preprocessed sensor data include peak value, RMS value, root square amplitude and kurtosis; the frequency-domain spectral features include main frequency amplitude, sideband energy and spectral centroid; and the time-frequency domain features include wavelet packet node energy.

[0026] In one specific embodiment, continuous operating data is divided into operating stages with different process loads based on production formula parameters and equipment start / stop signals. The specific method is as follows: the starting point of each operating stage is triggered by the equipment start signal or the formula parameter switching signal, and the ending point is determined by the equipment stop signal or the next formula switching signal. Based on the key process parameters set in the formula of the current stage, a level label representing the load intensity is automatically calculated and assigned to the stage, thereby associating the operating data with the specific production operation status.

[0027] It should be noted that among the extracted time-domain statistical features, peak value and kurtosis are sensitive to the instantaneous impact components in the signal and can be used to indicate local damage such as early pitting of bearings or gears. The root mean square value characterizes the average energy level of the signal and is related to the overall vibration intensity of the equipment. In the frequency domain spectral features, the change in the amplitude of the main frequency can reflect faults such as rotor imbalance, while the sideband energy directly corresponds to the intensity of modulation phenomena caused by gear faults or bearing damage. The wavelet packet node energy features in the time and frequency domain are suitable for analyzing non-stationary signals and can effectively capture the energy changes of specific frequency bands in vibration signals, thereby identifying faults under variable speed operation. The purpose of demodulating and analyzing the current signal is to separate the low-frequency components modulated by the motor power supply frequency, which reflect the torque fluctuations or fault characteristics on the mechanical side. This is often achieved by methods such as Hilbert transform. The sideband energy features obtained after demodulation are key indicators for diagnosing early faults in mechanical transmission systems. The torque pulsation feature directly characterizes the smoothness of the transmission chain output. Its increase may indicate misalignment or abnormal load connection. The instantaneous efficiency feature reflects the health status of the drive system from an energy perspective. An abnormal decrease in efficiency may indicate an increase in electrical or mechanical losses.

[0028] In a specific embodiment of the present invention, the method for generating the feature index sequence reflecting the state of the target manufacturing equipment is as follows: the time-domain statistical features, frequency-domain spectrum features, time-frequency domain features, current harmonic features, sideband features, torque pulsation features, instantaneous efficiency features, and statistical features characterizing load intensity and working mode extracted within the same time window are combined in a preset dimensional order to form a unified multidimensional feature vector, and the multidimensional feature vector generated in time order is arranged to form the feature index sequence reflecting the state of the target manufacturing equipment.

[0029] For example, the time window is 10 seconds, 30 seconds, 40 seconds, etc.

[0030] It should be noted that in the process of generating the feature index sequence, all heterogeneous features extracted within the same time window are combined and spliced ​​according to a preset, fixed dimensional order to form a multi-dimensional state feature vector representing the overall state of the equipment at that moment. Each dimension of this vector corresponds to a specific physical or technological feature, and its structure remains consistent throughout the entire sequence. According to the order of each time window, these state feature vectors are arranged into a continuous time series, which constitutes the final feature index sequence. This data structure realizes the standardized encapsulation and spatiotemporal alignment of multi-source heterogeneous data. Its core value lies in converting discrete monitoring data of different properties into a state evolution sequence with a unified structure, time index, and usable for time series analysis. This provides a standardized input for subsequent fault prediction models that can reflect both instantaneous state and characterize evolution laws.

[0031] The intelligent diagnosis and early warning module is used to input the feature index sequence into a pre-trained multi-fault mode recognition model, calculate the potential fault probability and early warning level of the target manufacturing equipment, and generate early warning instructions.

[0032] In a specific embodiment of the present invention, the calculation method for calculating the potential failure probability and early warning level of the target manufacturing equipment is as follows: inputting the feature index sequence into a pre-trained multi-fault mode recognition model, and outputting the probability of the target manufacturing equipment experiencing multiple predefined fault types, including bearing wear, gear damage, shaft misalignment, mechanical loosening, and electrical insulation aging.

[0033] For each fault type, its corresponding fault occurrence probability is compared with multiple preset graded probability thresholds for that fault type, and it is mapped to the corresponding warning level according to the graded probability threshold range in which the fault occurrence probability is located.

[0034] It should be noted that the fault types are associated with characteristic index sequences. For example, bearing wear faults are mainly associated with time-domain statistical characteristics, frequency-domain spectrum characteristics, and sideband characteristics of current signals. Gear damage faults are mainly associated with frequency-domain spectrum characteristics of vibration signals and current harmonic characteristics and sideband characteristics of current signals.

[0035] It should be noted that the graded probability threshold for each type of fault is not a fixed value, but is determined based on the characteristic distribution of the fault in the historical data of normal equipment operation, the physical model of fault evolution, and actual maintenance experience. For example, the distribution of the probability values ​​output by the model under historical normal conditions and known fault conditions can be statistically analyzed to set threshold ranges that distinguish different risk levels, and support adaptive calibration of the threshold based on the operating data of specific equipment.

[0036] It should be noted that mapping probability values ​​to discrete warning levels, such as Level 1, Level 2, and Level 3, aims to transform continuous probability values, which are not easily interpreted directly, into risk level signals that are directly linked to maintenance decisions. Different warning levels correspond to different response times, inspection depths, and handling priorities. For example, a Level 1 warning triggers periodic monitoring logs, a Level 2 warning triggers on-site inspection notifications, and a Level 3 warning triggers the preventive maintenance work order generation process, thereby seamlessly integrating the prediction results into the existing maintenance management system.

[0037] For example, assuming the model outputs a bearing wear failure probability of 0.7, according to the preset graded probability thresholds for bearing wear failure: when the failure probability is less than 0.3, it corresponds to a level 1 warning; when the failure probability is between 0.3 and 0.6, it corresponds to a level 2 warning; when the failure probability is between 0.6 and 0.8, it corresponds to a level 3 warning; and when the failure probability is greater than or equal to 0.8, it corresponds to a level 4 warning. After comparison, the failure probability of 0.7 falls within the threshold range of 0.6 to 0.8, so the bearing wear failure is mapped to a level 3 warning.

[0038] For example, assuming the model outputs a fault occurrence probability of 0.25 for electrical insulation aging faults, according to the preset graded probability thresholds for electrical insulation aging faults: when the fault occurrence probability is less than 0.2, it corresponds to a level 1 warning; when the fault occurrence probability is between 0.2 and 0.5, it corresponds to a level 2 warning; and when the fault occurrence probability is greater than or equal to 0.5, it corresponds to a level 3 warning. After comparison, the fault occurrence probability of 0.25 is within the threshold range of 0.2 to 0.5, so the electrical insulation aging fault is mapped to a level 2 warning.

[0039] In a specific embodiment of the present invention, the pre-trained multi-fault mode recognition model is specifically trained as follows: based on historical operating data of the target manufacturing equipment under normal operating conditions and fault type conditions, a training sample set corresponding to fault type labels is constructed. The training sample set is used to perform supervised training on a multi-task classification model constructed based on a deep neural network. The output layer of the multi-task classification model consists of multiple parallel branches, each branch corresponding to a predefined fault type and outputting the probability value of the occurrence of that type of fault. During the training process, the predicted output of each branch is calculated through forward propagation. The cross-entropy loss function is used to quantify the difference between the predicted probability distribution and the true label distribution. The model parameters are iteratively updated based on the backpropagation algorithm to minimize the loss function, thereby obtaining a multi-fault mode recognition model.

[0040] It should be noted that the fault types in the training sample set for constructing the corresponding fault type labels correspond to the predefined fault types. The labels of the training sample set are generated based on actual fault records or experimentally simulated fault data in historical data, aiming to cover the predefined set of fault types to ensure that the trained model has the ability to identify these target faults.

[0041] It should be noted that the training of the multi-task classification model adopts the conventional supervised learning process in the field of deep learning, including steps such as calculating the network output through forward propagation, evaluating the prediction error using the cross-entropy loss function, and updating the network weights based on the backpropagation algorithm. The above methods are common techniques for training deep neural networks in this field, and their specific implementations can be found in relevant known technologies, which will not be detailed here.

[0042] In a specific embodiment of the present invention, the generation of the early warning instruction specifically includes: generating a structured instruction containing the fault type, early warning level, timestamp, and corresponding visual marker information according to the early warning level of each fault type. The visual marker information is assigned different shape markers according to the fault type and different color markers according to its early warning level.

[0043] For example, visual labeling information can be used, such as assigning a circular label to bearing wear faults, a triangular label to gear damage faults, and a square label to shaft misalignment faults. At the same time, a green label is assigned to a first-level warning, a yellow label to a second-level warning, and a red label to a third-level warning.

[0044] The maintenance decision generation module is used to call the digital twin model corresponding to the target manufacturing equipment to simulate the fault evolution and maintenance process, and output a predictive maintenance plan for the target manufacturing equipment.

[0045] In a specific embodiment of the present invention, the method of calling the digital twin model corresponding to the target manufacturing equipment to simulate fault evolution and maintenance process is as follows: according to the fault type and warning level in the warning instruction, the corresponding characteristic index sequence is input into the digital twin model; the trend of the fault type in the future operating cycle is simulated according to the constructed digital twin model; according to the current fault type and its warning level, the corresponding maintenance strategy is matched from the maintenance strategy corresponding to the fault type and its warning level stored in the local database; the complete maintenance operation process for the current fault type is simulated and executed; and the simulated maintenance time, resource consumption list and post-maintenance performance recovery parameters under the maintenance strategy are output.

[0046] It should be noted that the trend of the fault type calculated by the driving twin model in the future operating cycle refers to the fault state reflected by the current characteristic index sequence. It is the dynamic change process of the key performance indicators of the fault under continuous operating conditions, which is deduced through physical simulation or data-driven model. For example, for bearing wear faults, the trend can be shown as a curve in which the effective value of vibration acceleration gradually increases over time. For electrical insulation aging faults, the trend can be shown as a curve in which the leakage current of the motor winding increases with the increase of temperature. This trend is used to quantify the evolution speed and severity of the fault, and to provide a time basis for the formulation of subsequent maintenance strategies.

[0047] For example, the maintenance strategies corresponding to the fault type and its warning level are as follows: for a level three warning of bearing wear fault, the matching maintenance strategy is to replace the main shaft bearing and re-lubricate it; for a level two warning of gear damage fault, the matching strategy is to disassemble and inspect the gearbox and repair or replace the damaged gear pair; for a level one warning of shaft misalignment fault, the matching strategy is to use laser alignment adjustment and tighten the anchor bolts; for a level three warning of electrical insulation aging fault, the matching strategy is to perform local insulation reinforcement treatment or replace the stator winding in sections.

[0048] It should be noted that the resource consumption list refers to the detailed list of all resources required to complete the maintenance operation when the maintenance strategy simulation is executed in the digital twin model. This includes, but is not limited to, the model and quantity of spare parts required, the estimated man-hours, the time occupied by special tools or equipment, and the auxiliary materials that may be consumed. The performance recovery parameters after maintenance refer to the simulated evaluation values ​​of the key performance indicators of the equipment after the maintenance operation is completed in the simulation model, such as: the motor energy efficiency is restored to more than 95% of the rated value or the machining accuracy is restored to the process tolerance range.

[0049] In a specific embodiment of the present invention, the digital twin model is constructed as follows: a three-dimensional geometric model of the target manufacturing equipment is constructed based on its design model and material properties, and a multi-domain physical simulation model is established based on the kinematic principles, material mechanical properties and electrical control logic of the target manufacturing equipment. By defining a unified spatiotemporal reference, the geometric model, the physical simulation model and the real-time sensor data, driver operation data and process control system data of the target manufacturing equipment are dynamically bound and synchronized to obtain the digital twin model.

[0050] It should be noted that, in this invention, kinematic principles specifically refer to the relative motion constraints between the moving parts of the target manufacturing equipment, and the resulting changes in motion parameters such as part position, velocity, and acceleration over time. Material mechanical properties specifically refer to the physical properties of the materials constituting the key components of the equipment, as well as the overall structural stiffness, mass distribution, and dynamic response characteristics determined by the geometry of the components. Electrical control logic specifically refers to the sequential control and interlocking protection programs executed in the programmable logic controller of the equipment, and the speed loop, position loop, and current loop adjustment algorithms and parameters implemented inside the servo drive. These principles and characteristics are the core knowledge foundation upon which a high-fidelity digital twin simulation model that can accurately reflect the actual physical behavior of the equipment is constructed.

[0051] In one specific embodiment, the geometric model of the target manufacturing equipment is constructed based on the design model and material properties. The specific method is as follows: according to the computer-aided design drawings of the target manufacturing equipment, a three-dimensional geometric model containing the shape, assembly relationship and spatial position of all key mechanical components is established through professional modeling software, and the material property parameters of each component are added to the corresponding part of the model to construct its three-dimensional geometric model.

[0052] In one specific embodiment, a multi-domain physical simulation model of the target manufacturing equipment is established based on its kinematic principles, material mechanical properties, and electrical control logic. The specific method is as follows: on the basis of the three-dimensional geometric model, according to the actual working principle of the target manufacturing equipment, a multi-body dynamics model of the motion relationship of its movable parts, a finite element analysis model reflecting the stress and strain distribution of structural components under load, a thermodynamic model simulating the heat exchange process, and an electrical and control simulation model characterizing the behavior of the motor, driver, and control system are constructed. By defining the coupling boundary conditions and energy transfer paths between the models, these simulation models belonging to different physical domains are integrated into a unified overall simulation system that can comprehensively reflect the dynamic behavior of the equipment's mechanical-electrical-thermal multi-field coupling.

[0053] It should be noted that dynamically binding and synchronizing the geometric model, physical simulation model, and real-time sensor data, actuator operation data, and process control system data of the target manufacturing equipment by defining a unified spatiotemporal reference means setting the same clock source for the digital twin model and the target manufacturing equipment, and using a fixed mechanical position or electrical signal cycle of the target manufacturing equipment as the synchronization origin in space and time. Through an industrial communication interface, the real-time sensor data, actuator operation data, and process control system data are transmitted to the digital twin model. After receiving the data, the digital twin model drives the corresponding components in the three-dimensional geometric model to move and updates the boundary conditions and input parameters of the physical simulation model, so that the state of the digital twin model is consistent with the real-time operating state of the physical equipment, thus achieving virtual-real synchronization.

[0054] In a specific embodiment of the present invention, the method for outputting the predictive maintenance scheme for the target manufacturing equipment is as follows: Based on the received early warning instructions and their corresponding digital twin simulation results, a structured executable maintenance scheme is generated. For each early warning fault, the execution priority and order of each early warning fault are determined according to its early warning level, the maintenance operation duration simulated by the digital twin model, the resource consumption list, and in combination with the production plan and maintenance resource status of the target manufacturing equipment. The early warning faults are then sorted by time sequence and resource allocation, and a predictive maintenance work order is output, which includes the maintenance task sequence, planned time, resource arrangement, and expected performance indicators. Simultaneously, an analysis report with visualization markers and trend curves is generated.

[0055] It should be noted that, based on the received warning commands and their corresponding digital twin simulation results, a structured and executable maintenance plan is generated. This step refers to automatically associating and integrating the warning commands with the simulation results output by the digital twin module to form a maintenance plan that can be directly executed. For example, when the system receives a command for spindle bearing wear and a level 3 warning, it calls the maintenance strategy that has been simulated for this fault: replace the bearing and relubricate, the simulation time is 4.5 hours, and the required resource list includes two bearing models XYZ, 0.1 liters of grease, and one set of special puller. This information is structured and organized under the same plan item.

[0056] It should be noted that determining the execution priority and sequence of various early warning faults, and prioritizing their timing and resource allocation, involves using a predefined decision-making logic. This logic comprehensively considers the early warning level, simulated repair time, and resource consumption requirements of each fault, while also integrating the target equipment's production plan and real-time maintenance resource status. The core principles of this decision-making logic include: faults with higher early warning levels are assigned higher execution priority; tasks with longer simulated repair times are allocated suitable, earlier available time windows to minimize potential impact on subsequent production plans; and tasks are prioritized for matching with available resources and ensuring production continuity. The time period with the least impact, for example: there are two faults to be dealt with: Fault A, bearing wear, level 3 warning, simulated repair time 4.5 hours, and the required bearing inventory is sufficient; and Fault B, gear damage, level 2 warning, simulated repair time 2 hours, and the required gear inventory will arrive in 2 days. Considering the production plan and resource status for the next 24 hours, there is a continuous 6-hour maintenance window in the production plan. The decision is as follows: Fault A is given the highest priority and is scheduled to be executed first in this window because it has a higher warning level, the required resources are available immediately, and its longer repair time can fully match the 6-hour window. Fault B is automatically given a lower priority because of the shortage of required spare parts. The sorting decision is to postpone it to the next available window after the spare parts arrive, and the spare parts procurement prompt process is triggered simultaneously.

[0057] It should be noted that outputting predictive maintenance work orders, which include maintenance task sequences, planned times, resource allocation, and expected performance indicators, and simultaneously generating an analysis report with visual markers and trend curves, refers to formatting the aforementioned sorting and allocation results into standard work orders and auxiliary analysis documents. The output work orders will be clearly listed, such as: Task 1: Replace spindle bearing, planned time: May 20th, 18:00-22:30, required resources: 2 bearings XYZ, shift A, expected performance: vibration value ≤2.0mm / s. The simultaneously generated analysis report will enhance the information presentation in a visual way. For example, on the equipment layout diagram, a red circle will be marked at the spindle position to indicate a bearing wear fault with a level 3 warning, and a trend curve will be displayed next to it from the historical vibration value simulation prediction to the recovery value after repair. This report, combined with the structural work order, provides clear instructions for on-site maintenance and provides intuitive status retrospective and effect prediction basis for management decisions.

[0058] Reference Figure 2As shown, the present invention provides a method for predictive maintenance system of intelligent manufacturing equipment, including step 1. Multi-source data processing and feature construction: sensor data, driver operation data and process control system data of the target manufacturing equipment are obtained through standardized interfaces, and data preprocessing and feature extraction are performed to generate a feature index sequence reflecting the status of the target manufacturing equipment.

[0059] Step 2. Intelligent diagnosis and early warning: Input the feature index sequence into the pre-trained multi-fault mode recognition model, calculate the potential fault probability and early warning level of the target manufacturing equipment, and generate early warning instructions.

[0060] Step 3. Maintenance Decision Generation: Call the digital twin model corresponding to the target manufacturing equipment to simulate the fault evolution and maintenance process, and output a predictive maintenance plan for the target manufacturing equipment.

[0061] The examples described in this invention are not limited to the specific embodiments listed above. The examples are merely illustrative to facilitate understanding of the invention and do not constitute a limitation on the scope of protection of this invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of this invention should be included within the scope of protection.

[0062] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. A predictive maintenance system for intelligent manufacturing equipment, characterized in that, Includes the following modules: Multi-source data processing and feature construction module: used to acquire sensor data, driver operation data and process control system data of target manufacturing equipment through standardized interfaces, and perform data preprocessing and feature extraction to generate a sequence of feature indicators reflecting the status of target manufacturing equipment; Intelligent diagnosis and early warning module: used to input the feature index sequence into a pre-trained multi-fault mode recognition model, calculate the potential fault probability and early warning level of the target manufacturing equipment, and generate early warning instructions; Maintenance Decision Generation Module: This module is used to call upon the digital twin model corresponding to the target manufacturing equipment to simulate fault evolution and maintenance process, and output predictive maintenance solutions for the target manufacturing equipment.

2. The predictive maintenance system for intelligent manufacturing equipment according to claim 1, characterized in that, The specific methods for data preprocessing and feature extraction are as follows: Wavelet denoising, bandpass filtering, and outlier removal are performed on the acquired sensor data and driver operation data. Time-domain statistical features, frequency-domain spectral features, and time-frequency features are extracted from the preprocessed sensor data. Current harmonic features and sideband features characterizing the mechanical transmission state are extracted from the preprocessed driver operation data by demodulating and analyzing the current signal. Torque pulsation features are extracted from the speed and torque signals, and instantaneous efficiency features are extracted based on electrical parameters. The process control system data is synchronized and aligned based on the master clock timestamp, and the continuous operation data is divided into operating stages with different process loads according to the production formula parameters and equipment start-stop signals. Statistical features are extracted from the preprocessed and segmented process control system data to characterize the load intensity and working mode of the target manufacturing equipment under each operating stage.

3. The predictive maintenance system for intelligent manufacturing equipment according to claim 2, characterized in that, The specific method for generating the feature index sequence reflecting the state of the target manufacturing equipment is as follows: The time-domain statistical features, frequency-domain spectral features, time-frequency domain features, current harmonic features, sideband features, torque pulsation features, instantaneous efficiency features, and statistical features characterizing load intensity and operating mode extracted within the same time window are combined in a preset dimensional order to form a unified multidimensional feature vector. The multidimensional feature vectors generated in time order are then arranged to form a sequence of feature indicators reflecting the status of the target manufacturing equipment.

4. The predictive maintenance system for intelligent manufacturing equipment according to claim 3, characterized in that, The specific calculation method for the potential failure probability and early warning level of the target manufacturing equipment is as follows: The feature index sequence is input into a pre-trained multi-fault mode recognition model, which outputs the probability of various predefined fault types occurring in the target manufacturing equipment. The fault types include bearing wear, gear damage, shaft misalignment, mechanical loosening, and electrical insulation aging. For each fault type, its corresponding fault occurrence probability is compared with multiple preset graded probability thresholds for that fault type, and it is mapped to the corresponding warning level according to the graded probability threshold range in which the fault occurrence probability is located.

5. The predictive maintenance system for intelligent manufacturing equipment according to claim 4, characterized in that, The specific training method for the pre-trained multi-fault mode recognition model is as follows: Based on historical operating data of the target manufacturing equipment under normal operating and fault type conditions, a training sample set corresponding to the fault type labels is constructed. The training sample set is used to conduct supervised training on a multi-task classification model based on a deep neural network. The output layer of the multi-task classification model consists of multiple parallel branches, each branch corresponding to a predefined fault type and outputting the probability value of the occurrence of that type of fault. During the training process, the predicted output of each branch is calculated through forward propagation. The cross-entropy loss function is used to quantify the difference between the predicted probability distribution and the true label distribution. The model parameters are iteratively updated based on the backpropagation algorithm to minimize the loss function, thereby obtaining multiple fault mode recognition models.

6. The predictive maintenance system for intelligent manufacturing equipment according to claim 3, characterized in that, The specific content of the generated early warning instruction is as follows: Based on the warning level of each fault type, a structured instruction is generated that includes the fault type, warning level, timestamp, and corresponding visual marker information. The visual marker information is assigned different shape markers according to the fault type and different color markers according to its warning level.

7. The predictive maintenance system for intelligent manufacturing equipment according to claim 6, characterized in that, The specific method for calling the digital twin model corresponding to the target manufacturing equipment to simulate fault evolution and maintenance process is as follows: Based on the fault type and warning level in the warning instruction, the corresponding characteristic index sequence is input into the digital twin model. The constructed digital twin model simulates the trend of the fault type in the future operating cycle. Based on the current fault type and its warning level, the corresponding maintenance strategy is matched from the maintenance strategy corresponding to the fault type and its warning level stored in the local database. The complete maintenance operation process for the current fault type is simulated and executed. The simulated maintenance time, resource consumption list and post-maintenance performance recovery parameters under the maintenance strategy are output.

8. The predictive maintenance system for intelligent manufacturing equipment according to claim 7, characterized in that, The specific construction method of the constructed digital twin model is as follows: A three-dimensional geometric model of the target manufacturing equipment is constructed based on its design model and material properties. A multi-domain physical simulation model is established based on the kinematic principles, material mechanical properties and electrical control logic of the target manufacturing equipment. By defining a unified spatiotemporal reference, the geometric model, physical simulation model and the real-time sensor data, drive operation data and process control system data of the target manufacturing equipment are dynamically bound and synchronized to obtain a digital twin model.

9. The predictive maintenance system for intelligent manufacturing equipment according to claim 7, characterized in that, The specific method for the predictive maintenance scheme of the output target manufacturing equipment is as follows: Based on the received early warning instructions and their corresponding digital twin simulation results, a structured and executable maintenance plan is generated. For each early warning fault, the execution priority and sequence of each early warning fault are determined according to its early warning level, the maintenance operation duration simulated by the digital twin model, the resource consumption list, and in combination with the production plan and maintenance resource status of the target manufacturing equipment. The early warning faults are sorted by time sequence and resource allocation, and a predictive maintenance work order is output, which includes the maintenance task sequence, planned time, resource arrangement, and expected performance indicators. An analysis report with visualization markers and trend curves is generated simultaneously.

10. A method for implementing a predictive maintenance system for intelligent manufacturing equipment according to any one of claims 1-9, characterized in that, include: Step 1. Multi-source data processing and feature construction: Sensor data, actuator operation data and process control system data of the target manufacturing equipment are obtained through standardized interfaces, and data preprocessing and feature extraction are performed to generate a sequence of feature indicators reflecting the status of the target manufacturing equipment. Step 2. Intelligent diagnosis and early warning: Input the feature index sequence into the pre-trained multi-fault mode recognition model, calculate the potential fault probability and early warning level of the target manufacturing equipment, and generate early warning instructions; Step 3. Maintenance Decision Generation: Call the digital twin model corresponding to the target manufacturing equipment to simulate the fault evolution and maintenance process, and output a predictive maintenance plan for the target manufacturing equipment.

Images