Equipment intelligent predictive maintenance system fusing production scheduling

Abstract

The invention discloses an equipment intelligent predictive maintenance system fusing production scheduling, relates to the technical field of industrial data processing and product life cycle management, and aims to solve the problems of decision conflicts and quality risks caused by separation of traditional predictive maintenance and a production plan. The system comprises an equipment prediction module which is used for generating a future decline trend curve of each dimension by adopting a time sequence prediction model and triggering early warning by calculating failure momentum representing a degradation rate so as to realize transformation from fault prediction to process quality guarantee. The system further comprises a maintenance simulation analysis module which inserts maintenance tasks into different time windows through simulation, quantifies ripple effects generated by the maintenance tasks on a production plan, constructs a multi-dimensional comprehensive cost value including delivery, productivity, scheduling and the like, further introduces a carbon footprint consumption threshold value to perform scheme pruning, and finally, in a scheme meeting environmental protection constraints, the maintenance simulation analysis module performs maintenance simulation analysis on the production plan. And selecting one with the lowest comprehensive cost to generate a collaborative scheduling instruction.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing and product lifecycle management technology, specifically to an intelligent predictive maintenance system for equipment that integrates production scheduling. Background Technology

[0002] In the macro-strategy of modern manufacturing, Product Lifecycle Management (PLM) encompasses asset performance management, which oversees the operational status and health of physical assets such as production equipment. Stable operation of production equipment is the core foundation for ensuring product quality, production efficiency, and corporate economic benefits. To maximize equipment utilization and avoid significant losses caused by unplanned downtime, equipment maintenance strategies have evolved from passive "post-failure repair" and fixed "periodic maintenance" to the current mainstream and more advanced "condition-based predictive maintenance."

[0003] Predictive maintenance technology deploys various sensors to collect real-time equipment operating parameters (such as vibration, temperature, and current), and uses data analysis and algorithm models to predict potential equipment failures or performance degradation trends. This allows for early warnings before failures actually occur, guiding maintenance activities. Compared to traditional maintenance methods, predictive maintenance significantly improves the accuracy and foresight of maintenance, effectively reducing severe production interruptions caused by sudden equipment failures.

[0004] However, existing predictive maintenance technologies still face several deep-seated technical bottlenecks in practical applications: Traditional predictive maintenance systems operate as isolated information silos. When the system predicts an impending equipment failure and issues a maintenance alert, this information is directly transmitted to the maintenance department. However, the production department's scheduling system remains unaware of this information. This leads to a sharp conflict between maintenance needs and production tasks: immediate downtime for maintenance disrupts established production plans, causing order delays and capacity losses; while delaying maintenance may result in equipment failure during critical production tasks, causing even more severe losses. Decision-makers are forced to make trade-offs based solely on experience without quantitative evidence, making it difficult to find the globally optimal maintenance timing. Most predictive models focus on the overall health index of the equipment or the failure probability of specific components, and the mapping relationship between the monitored physical quantities and the final product processing quality is unclear. For example, equipment with an overall health level of 80% may not be able to meet the spatial positioning accuracy requirements of high-precision orders. This vague assessment method makes it impossible to accurately correlate maintenance alerts with their impact on specific production tasks, reducing the effectiveness of the alerts. When selecting maintenance windows, existing considerations are limited to "finding equipment idle time," lacking a systematic and quantitative assessment of the complex "ripple effects" caused by inserting maintenance tasks. Seemingly "idle" windows can generate huge, hidden overall costs by disrupting material flow, consuming key human resources, or causing a chain reaction of delays in subsequent processes. Meanwhile, the decision-making process often overlooks key indicators of sustainable manufacturing, such as energy consumption and carbon footprint. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent predictive maintenance system for equipment that integrates production scheduling, thereby solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent predictive maintenance system for equipment integrating production scheduling, comprising: The equipment prediction module is used to establish a structured data model for each piece of equipment in the production system, which is directly related to the skill dimensions of the equipment and the processing quality. It acquires the operating parameter data of the equipment in real time and converts the operating parameter data into quantitative index values ​​for each skill dimension in the skill profile. The quantitative index values ​​include, but are not limited to, spatial positioning accuracy index, spindle performance index, tool stability index and thermal stability index. The real-time generated quantitative index values ​​are recorded in chronological order to generate a historical quantitative index value sequence for each piece of equipment. A time series prediction model is used to generate a future decline trend curve for each skill dimension and generate the failure momentum of the i-th device in each dimension at future time t. A failure momentum upper limit threshold is preset. When the failure momentum of each skill dimension exceeds the failure momentum upper limit threshold, the first warning instruction is triggered. The maintenance simulation analysis module is used to execute a demand matching process when making equipment allocation decisions in production scheduling after receiving the first early warning instruction. It evaluates and obtains the first sorting queue and the second sorting queue. It filters in the first sorting queue and identifies and generates multiple potential maintenance execution time windows based on the task arrangement of the equipment in the production scheduling plan, forming the first potential maintenance execution time window column. The simulation scheduling schemes corresponding to the first potential maintenance execution time window column are analyzed to quantify the ripple effect caused by the insertion of maintenance tasks and construct the comprehensive cost value and total energy consumption value. A carbon footprint consumption threshold is preset, and the simulation scheduling schemes with total energy consumption values ​​exceeding the carbon footprint consumption threshold are eliminated. The potential maintenance execution time window with the smallest comprehensive cost value is selected as the optimal solution, and the first collaborative scheduling instruction is generated.

[0007] Preferably, the equipment prediction module includes an equipment skill profile building unit and a failure prediction unit; The equipment prediction module includes an equipment skill profile building unit and a failure prediction unit; The skill profile building unit is used to build a skill profile for each device, create a digital twin skill model for each device, acquire the device's operating parameter data in real time, and conduct training and verification to build the spatial positioning accuracy index, spindle performance index, tool stability index and thermal stability index for each device. The failure prediction unit is used to generate a sequence of historical quantitative index values ​​based on the spatial positioning accuracy index, spindle performance index, tool stability index, and thermal stability index of each device. It uses LSTM neural network technology to build a time series prediction model, trains the historical quantitative index value sequence, plots the future decline trend curve for each skill dimension, and generates the failure momentum of each dimension of the i-th device at future time t. The decline trend curve consists of a set of multi-dimensional core performance index prediction values ​​for future time t, covering spatial positioning accuracy, spindle performance, tool stability, and thermal stability. The method for obtaining the failure momentum of each skill dimension in the i-th device at future time t is as follows: Select a future time window, from the current time tn to the future time tn+k after k time steps, to obtain the first future time window. For each skill dimension, use linear regression to fit the predicted index value sequence within the first future time window, and extract the slope of the fitted line as the failure momentum of that skill dimension. The length of the first future time window is set based on the equipment's historical mean time between failures (MTBF). A preset upper limit threshold for failure momentum is set. When the failure momentum of each skill dimension exceeds the upper limit threshold, the first warning instruction is triggered.

[0008] Preferably, the spatial positioning accuracy index is obtained by acquiring the servo motor command position signal and the actual position signal fed back by the grating ruler or encoder, calculating the dynamic position error between the two, obtaining an error sequence, extracting the maximum overshoot, root mean square error, and settling time that characterize the positioning performance degradation from the error sequence, and normalizing them according to the preset thresholds corresponding to the maximum overshoot, root mean square error, and settling time to generate independent, dimensionless sub-item positioning degradation degrees. The sub-item positioning degradation degrees are then linearly combined using preset weighting coefficients to generate a composite positioning degradation degree. The composite positioning degradation degree is negatively correlated with the spatial positioning accuracy index, which is measured with a maximum score of 100. The final score is obtained by subtracting a loss factor proportional to the magnitude of the composite positioning degradation degree from the maximum score.

[0009] Preferably, the spindle performance index is obtained by acquiring the spindle vibration signal of the equipment and performing a fast Fourier transform on the spindle vibration signal to obtain the vibration spectrum; From the vibration spectrum, characteristic frequencies corresponding to the health status of key bearing components are identified and extracted, specifically including the characteristic frequencies of the bearing outer ring, inner ring, and rolling elements, and their corresponding energy amplitudes are obtained. The energy amplitudes of the outer ring, inner ring, and rolling elements are normalized according to their respective preset threshold values ​​to generate independent, dimensionless sub-items of vibration degradation. The sub-items of vibration degradation are linearly combined according to preset weighting coefficients that sum to 1 to generate a comprehensive vibration degradation degree that fully reflects the overall health status of the bearing. The spindle performance index is measured with a maximum score of 100, and its final score is obtained by subtracting a loss factor proportional to the magnitude of the comprehensive vibration degradation degree from the maximum score.

[0010] Preferably, the tool stability index is obtained by using the tool clamping or releasing command as the starting trigger signal for data acquisition when executing the tool change command, and using the signal of the corresponding travel limit switch being triggered or the pressure reaching a stable state as the termination signal, thereby capturing a first future time window. Within the first future time window, the output value of the pressure sensor installed in the hydraulic or pneumatic actuation circuit of the clamping or releasing mechanism is acquired at high frequency to form a dynamic pressure curve that can characterize the entire process of each clamping or releasing action. The dynamic pressure curve is time-aligned with the reference pressure representing the healthy state that is pre-stored in the system, and the cumulative waveform distance between the two is calculated. At the same time, the deviation value between the actual duration of this action and the reference time is calculated to form the time deviation. The cumulative distance and time deviation of the waveform are normalized according to their preset thresholds to generate distance degradation and time degradation. The distance degradation and time degradation are weighted and combined to form a comprehensive offset metric. The tool stability index is measured with 100 points as the full score, and its final score is obtained by deducting the loss component proportional to the size of the comprehensive offset metric from the full score. The thermal stability index is obtained by simultaneously collecting the temperature change time series of at least one key heat source of the equipment and the thermally induced displacement time series of key reference points of the machine tool. Through correlation analysis, the thermal sensitivity coefficient K is calculated. The thermal sensitivity coefficient K quantifies the displacement caused by a unit temperature change, representing the thermal response characteristics of the equipment. The thermal sensitivity coefficients in multiple working cycles are extracted to obtain a set of sensitivity coefficients, and the standard deviation of the set of sensitivity coefficients is calculated. The standard deviation is normalized according to its preset performance threshold to generate a dimensionless thermal instability. The thermal stability index is measured with a full score of 100. Its final score is obtained by deducting a loss factor proportional to the magnitude of the thermal instability from the full score.

[0011] Preferably, the maintenance simulation analysis module includes a demand matching unit; The demand matching unit is used to automatically perform demand matching in response to the first warning command; When making equipment allocation decisions for production scheduling, a demand matching process is executed, and a lower limit threshold for failure momentum is preset. When the failure momentum of the i-th equipment in each dimension is lower than the lower limit threshold for failure momentum at future time t, it means that the equipment is qualified in the current skill dimension. The difference between the two is calculated to obtain the health margin. Based on the corresponding process of each skill dimension, the health margins are sorted from high to low to obtain the first sorting queue. The system continuously monitors the failure momentum of all devices. When the failure momentum of any device in any skill dimension exceeds the preset upper limit threshold, a first warning instruction is triggered. The first warning instruction includes maintenance data related to the device. The system calculates the excess value of the failure momentum exceeding the upper limit threshold and defines the excess value as the maintenance urgency index of the current device. Based on the maintenance urgency index, all devices that have triggered warnings are sorted from high to low to form a second sorting queue. According to the second sorting queue, the initial priority of maintenance tasks is obtained. The initial priority of maintenance tasks is positively correlated with the maintenance urgency index, that is, the higher the maintenance urgency index, the higher the initial priority is set, and the higher the resource allocation weight is enjoyed when matching with maintenance resource data. Maintenance resource data includes: available work schedules for each member of the maintenance team, real-time inventory and estimated delivery time of required spare parts, and non-production windows in the production plan where maintenance tasks can be inserted.

[0012] Preferably, when the second sorting queue is not empty, the first maintenance planning process is activated according to the candidate devices in the second sorting queue, specifically as follows: Based on the task arrangement of the current candidate equipment in the production schedule, the task dependency of the current equipment is identified. The task dependency is determined based on the production task priority in the production schedule. The task dependency is obtained by adding the scores of the directly downstream tasks related to the equipment. If the candidate equipment is assigned a high-priority production task, then that equipment will be temporarily skipped, and maintenance planning will be prioritized for the next equipment in the second sorting queue that is not assigned a high-priority task.

[0013] Preferably, the maintenance simulation analysis module further includes a simulation scheduling scheme unit and a comprehensive cost analysis unit. The simulation scheduling scheme unit is used to filter the equipment to be planned based on the first sorting queue, identify and generate multiple potential maintenance execution time windows based on the task arrangement of the equipment in the production scheduling plan, and perform an independent scheduling recalculation simulation for each potential maintenance execution time window to generate a corresponding simulation scheduling scheme. Multiple potential maintenance execution time windows are used to generate multiple simulated scheduling schemes, forming the first potential maintenance execution time window column; The comprehensive cost analysis unit is used to analyze the corresponding simulated scheduling schemes in the first potential maintenance execution time window column to quantify the ripple effect caused by the insertion of maintenance tasks and construct the comprehensive cost value of the j-th corresponding simulated scheduling scheme for the i-th device.

[0014] Preferably, the method for obtaining the comprehensive cost value of the j-th corresponding simulated scheduling scheme for the i-th device is as follows: To calculate the cost of delayed delivery, identify all orders with delivery time delays in the simulated scheduling scheme, calculate the delay duration of each order, multiply it by a dynamic delay penalty coefficient that is positively correlated with the current order priority or customer level, and finally sum the calculation results of all orders to obtain the cost of delayed delivery. Calculate the cost of lost capacity by calculating the equipment downtime caused by the insertion of maintenance tasks, and multiply it by the unit time capacity value of the equipment. Calculate the indirect rescheduling cost, and calculate the additional material transfer time, new equipment process preparation time, and processing time increment caused by the efficiency difference of the alternative equipment when the affected process is moved to other alternative equipment to avoid delays. Convert these times into costs. To calculate supply chain disruption costs, analyze the changes in the completion time of all processes in the simulated scheduling scheme, quantify the advance or delay of the input material supply time of downstream processes, and calculate the cost by multiplying the amount of time drift by the supply chain stability penalty coefficient. Downstream processes include assembly or testing processes. The comprehensive cost value is obtained by directly adding the costs of delayed delivery, capacity loss, indirect rescheduling, and supply chain disruption.

[0015] Preferably, the maintenance simulation analysis module also includes a candidate window pruning unit and a cooperative scheduling instruction generation unit; The candidate window pruning unit is used to evaluate each simulated scheduling scheme to quantify the energy consumption effect caused by the insertion of maintenance tasks and construct the total energy consumption value of the j-th corresponding simulated scheduling scheme for the i-th device. The total energy consumption value of the i-th device in the j-th corresponding simulated scheduling scheme is obtained by multiplying the running time of each device in the corresponding simulated scheduling scheme by the power of the corresponding time period. Set a carbon footprint consumption threshold. When the total energy consumption of the j-th corresponding simulated scheduling scheme of the i-th device exceeds the carbon footprint consumption threshold, the current scheduling scheme will be removed from the potential maintenance execution time window to form a second potential maintenance execution time window column. The simulated scheduling schemes corresponding to the second potential maintenance execution time window column are traversed, and the potential maintenance execution time window with the smallest comprehensive cost value is selected as the optimal solution and the optimal maintenance intervention time window, and the first collaborative scheduling instruction is generated.

[0016] This invention provides an intelligent predictive maintenance system for equipment that integrates production scheduling. It offers the following advantages: (1) This invention achieves a fundamental shift from "predicting equipment failures" to "ensuring process quality," significantly improving the foresight and accuracy of maintenance decisions. Existing technologies mostly focus on the overall failure probability of equipment, and their prediction results are vaguely correlated with the processing quality of specific products. This invention innovatively constructs a "skill dimension" model (such as spatial positioning accuracy, spindle performance, etc.) directly linked to processing quality, and upgrades the prediction target from a static performance threshold to a dynamic "failure momentum." This method can keenly capture the early trend of accelerated degradation of equipment performance, and can provide early warnings even if its absolute performance value is still within the acceptable range. This makes maintenance intervention no longer a passive waiting for equipment to be on the verge of failure, but an active intervention before the equipment's processing capacity can meet specific process requirements, thereby effectively avoiding large-scale product quality problems caused by minor degradation of equipment performance, and elevating the core value of maintenance work from ensuring equipment operation to ensuring the quality of the final product.

[0017] (2) This invention constructs a globally optimal decision-making mechanism that unifies economic benefits and environmental responsibility, completely resolving the inherent conflict between maintenance activities and production plans. Existing technologies often rely solely on experience or a single "equipment idle" indicator when selecting maintenance timing, lacking a quantitative assessment of the complex impact on the entire production system. This invention, through a maintenance simulation analysis module, treats maintenance tasks as dynamic variables in production scheduling and innovatively quantifies the "ripple effect" generated by their implementation, constructing a "comprehensive cost value" encompassing multiple dimensions such as delivery, capacity, scheduling, and supply chain. Furthermore, this invention introduces a "carbon footprint consumption threshold" as a rigid constraint for decision-making, embedding sustainable development goals into the core of the algorithm. The resulting "coordinated scheduling instruction" is not a simple maintenance request, but an integrated solution that, after multi-scheme simulation and meeting environmental protection requirements, minimizes overall costs and includes specific time windows and optimized production plans. This achieves a leap from local, single-objective decision-making to global, multi-objective optimal decision-making, ensuring a high degree of synergy between corporate economic benefits and social responsibility. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system of the present invention; Figure 2 This is a schematic diagram of the system block diagram of the present invention; Figure 3 This is a schematic diagram of the system execution flow of the present invention. Detailed Implementation

[0019] 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.

[0020] Example 1 Please see Figures 1 to 3 This invention provides an intelligent predictive maintenance system for equipment that integrates production scheduling, comprising: The equipment prediction module is used to establish a structured data model for each piece of equipment in the production system, which is directly related to the skill dimensions of the equipment and the processing quality. It acquires the operating parameter data of the equipment in real time and converts the operating parameter data into quantitative index values ​​for each skill dimension in the skill profile. The quantitative index values ​​include, but are not limited to, spatial positioning accuracy index, spindle performance index, tool stability index and thermal stability index. The real-time generated quantitative index values ​​are recorded in chronological order to generate a historical quantitative index value sequence for each piece of equipment. A time series prediction model is used to generate a future decline trend curve for each skill dimension and generate the failure momentum of the i-th device in each dimension at future time t. A failure momentum upper limit threshold is preset. When the failure momentum of each skill dimension exceeds the failure momentum upper limit threshold, the first warning instruction is triggered. The maintenance simulation analysis module is used to execute a demand matching process when making equipment allocation decisions in production scheduling after receiving the first early warning instruction. It evaluates and obtains the first sorting queue and the second sorting queue. It filters in the first sorting queue and identifies and generates multiple potential maintenance execution time windows based on the task arrangement of the equipment in the production scheduling plan, forming the first potential maintenance execution time window column. The simulation scheduling schemes corresponding to the first potential maintenance execution time window column are analyzed to quantify the ripple effect caused by the insertion of maintenance tasks and construct the comprehensive cost value and total energy consumption value. A carbon footprint consumption threshold is preset, and the simulation scheduling schemes with total energy consumption values ​​exceeding the carbon footprint consumption threshold are eliminated. The potential maintenance execution time window with the smallest comprehensive cost value is selected as the optimal solution, and the first collaborative scheduling instruction is generated.

[0021] Figure 1 The isometric projection on the left illustrates the physical foundation of the system—a high-precision device within the production system. This corresponds to the starting point of the equipment prediction module: establishing a structured data model for the device that directly relates to processing quality in terms of skill dimensions, acquiring its operational parameter data in real time, and converting it into quantitative index values ​​for skill dimensions such as spatial positioning accuracy and spindle performance, generating a historical sequence of quantitative index values. The technical roadmap on the right reveals the core logic flow of this invention. The first flowchart, "Equipment Status Perception and Degradation Prediction," and the second flowchart, "Failure Momentum Analysis and Early Warning Trigger," together characterize the core function of the equipment prediction module: based on the historical sequence of quantitative index values, it uses a time series prediction model to generate future degradation trend curves, calculates the failure momentum in each dimension, and triggers a first early warning command when this momentum exceeds a preset upper limit threshold for failure momentum. The subsequent third flowchart, "Maintenance Window Simulation and Impact Quantification," and the fourth flowchart, "Multi-Objective Optimal Decision Making and Cooperative Scheduling," elaborate on the execution process of the maintenance simulation analysis module: After receiving the first warning instruction, the system identifies and generates multiple potential maintenance execution time windows based on the production scheduling plan. By analyzing the simulated scheduling scheme, the system quantifies the ripple effect caused by the insertion of maintenance tasks, constructs the comprehensive cost value and total energy consumption value, and finally, after eliminating schemes whose total energy consumption value exceeds the carbon footprint consumption threshold, selects the potential maintenance execution time window with the smallest comprehensive cost value as the optimal solution and generates the first cooperative scheduling instruction.

[0022] In this embodiment, compared with the prior art, the intelligent predictive maintenance system for equipment that integrates production scheduling provided by the present invention deeply couples and optimizes the refined prediction of equipment status with production scheduling decisions, specifically as follows: In existing technologies, maintenance early warning and production scheduling are two independent systems. Conflicts between them can only be resolved through manual coordination, which lacks quantitative basis and is inefficient. This invention, through a maintenance simulation analysis module, proactively integrates equipment maintenance needs as an endogenous variable into the dynamic planning of production scheduling. It does not simply issue alarms, but generates a "first collaborative scheduling instruction" that includes specific maintenance windows and an optimized production plan. This transforms passive conflict management into proactive collaborative optimization, fundamentally eliminating decision-making fragmentation and ensuring that maintenance activities are executed optimally while minimizing their impact on production, achieving global optimization of both production and maintenance objectives.

[0023] Traditional predictive models focus on the overall health of equipment, resulting in a vague correlation between the results and the quality requirements of specific processing tasks. This invention innovatively establishes a structured data model based on "skill dimensions" (such as spatial positioning accuracy and spindle performance), directly mapping equipment operating parameters to these quantitative indicators directly related to processing quality. By predicting the "failure momentum" of each skill dimension, the system can trigger an early warning before the equipment, still operational, fails to meet the processing accuracy requirements of a specific order. This refined predictive approach allows maintenance decisions to directly serve product quality assurance, effectively preventing the generation of large quantities of defective or scrap products due to minor equipment performance degradation, and significantly improving the foresight of quality control.

[0024] Existing technologies often only consider whether equipment is "idle" when selecting maintenance timing, ignoring the complex chain reactions it causes to the entire production system. This invention quantifies the "ripple effect" of inserting maintenance tasks, constructing a "comprehensive cost value" that includes delayed delivery costs, capacity loss costs, rescheduling costs, and supply chain disruption costs, transforming the complex system impact into clear, comparable numerical values. Furthermore, this invention introduces "total energy consumption" and "carbon footprint consumption threshold" as rigid constraints for decision-making, integrating sustainable manufacturing concepts into the core of the decision-making process. This makes the selection of the optimal maintenance window no longer a rough judgment based on experience, but a data-driven optimal decision based on a comprehensive and quantitative assessment of economic costs and environmental impact, greatly improving the scientific nature and overall benefits of the decision.

[0025] Example 2 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figures 1 to 3 Specifically, the equipment prediction module includes an equipment skill profile building unit and a failure prediction unit; The skill profile building unit is used to build a skill profile for each device. Specifically, it creates a digital twin skill model for each device based on the asset management shell AAS. This model is used to obtain the device's operating parameter data in real time for training and verification, in order to build the spatial positioning accuracy index, spindle performance index, tool stability index and thermal stability index for each device. The process of establishing a digital twin skill model involves inputting technical specifications of the equipment, CAD 3D models, electrical schematics, maintenance manuals, sensor lists, range, accuracy, and physical location information. Added sensor data: Data streams from externally deployed sensors are used to calibrate and verify operational hyperparameter data; The sensors used for spatial positioning accuracy index, spindle performance index, tool stability index, and thermal stability index include, but are not limited to: grating rulers / encoders, which are crucial for acquiring the "actual position signal" of the machine's motion axes (such as X, Y, and Z axes). Grating rulers are directly mounted between the moving parts and the machine bed, providing the most direct and accurate linear position feedback. Rotary encoders are mounted on the end of servo motors or ball screws, calculating linear displacement by measuring rotation angles. High-precision grating rulers are the preferred choice for ensuring the accuracy of these indices. CNC system: It needs to directly read the servo motor command position signal from the internal bus of the CNC system.

[0026] Accelerometers are standard sensors for measuring vibration. One or more high-frequency triaxial accelerometers are installed on the spindle housing near the front and rear bearings. This can comprehensively capture the characteristic high-frequency vibration signals generated by the spindle during rotation due to problems such as defects in the inner and outer rings of the bearings, rolling elements, or poor lubrication. High-frequency pressure sensor: This is crucial for monitoring this index. It needs to be installed in the hydraulic or pneumatic circuit of the tool clamping / releasing mechanism. The sensor must have a sufficiently high sampling rate to accurately capture the complete dynamic curve of pressure build-up, stabilization, and release within milliseconds. Travel limit switches: used to precisely define the start and end points of data acquisition. For example, the start signal is the "release" command issued by the control system, and the end signal is the triggering of the travel limit switch when the tool holder lever is detected to have moved into position. Temperature sensors are used when they need to be distributed across critical heat sources and structural points in equipment. Common types include thermocouples or resistance temperature detectors (RTDs). Deployment locations include: spindle motors, spindle bearings, ball screw nuts and support bearings, hydraulic stations, machine tool beds, and ambient temperature monitoring points.

[0027] High-precision displacement sensors are used to measure the "thermal-induced displacement" of critical reference points (such as the end of the spindle relative to the worktable). Because the displacement is at the micrometer level, non-contact, high-precision sensors are required, including eddy current sensors or laser displacement sensors.

[0028] Real-time acquisition of equipment operating parameter data. The source of operating parameter data is: data stream from the equipment PLC and CNC controller, such as servo axis position / speed / torque, spindle speed / load / current, G-code instructions, program running status, etc. (acquired through protocols such as OPCUA and MQTT).

[0029] After generating a standard-compliant AAS instance file (specifically in XML or JSON format) containing skill sub-models for each device, calculations are performed according to a preset algorithm to finally output the spatial positioning accuracy index, spindle performance index, tool stability index, and thermal stability index for each device.

[0030] The failure prediction unit is used to generate a sequence of historical quantitative index values ​​based on the spatial positioning accuracy index, spindle performance index, tool stability index, and thermal stability index of each device. It uses LSTM neural network technology to build a time series prediction model, trains the historical quantitative index value sequence, plots the future decline trend curve for each skill dimension, and generates the failure momentum of each dimension of the i-th device at future time t. The recession trend curve includes: A sequence of predicted spatial positioning accuracy index values ​​for future time t; A sequence of predicted spindle performance index values ​​at future time t; Predicted sequence of tool stability index values ​​at future time t; A sequence of predicted values ​​for the thermal stability index at future time t; The method for obtaining the failure momentum of each skill dimension in the i-th device at future time t is as follows: Select a future time window, from the current time tn to the future time tn+k after k time steps, to obtain the first future time window. For each skill dimension, use linear regression to fit the predicted index value sequence within the first future time window, and extract the slope of the fitted line as the failure momentum of that skill dimension. The length of the first future time window is set based on the equipment's historical mean time between failures (MTBF). A preset upper limit threshold for failure momentum is set. When the failure momentum of each skill dimension exceeds the upper limit threshold, the first warning instruction is triggered.

[0031] The spatial positioning accuracy index is obtained by acquiring the servo motor command position signal and the actual position signal fed back by the grating ruler or encoder of the device, calculating the dynamic position error between the two, obtaining an error sequence, extracting the maximum overshoot, root mean square error, and settling time that characterize the positioning performance degradation from the error sequence, and normalizing them according to the preset thresholds corresponding to the maximum overshoot, root mean square error, and settling time to generate independent, dimensionless sub-items of positioning degradation. The sub-items of positioning degradation are then linearly combined using preset weighting coefficients to generate a composite positioning degradation. The composite positioning degradation is negatively correlated with the spatial positioning accuracy index, which is measured with a maximum score of 100. The final score is obtained by subtracting a loss factor proportional to the magnitude of the composite positioning degradation from the maximum score.

[0032] First calculate the overshoot degradation. Error degradation and time degradation :

[0033] in, This is the actual measured value of the maximum overshoot. The preset maximum overshoot performance threshold, This is the actual measured value of the root mean square error. The preset root mean square error performance threshold, The actual measured value of the settling time. This is a preset stable time performance threshold; overshoot degradation Error degradation and time degradation After normalization, the composite location degradation degree is calculated using the following formula. :

[0034] in, , and These are the overshoot degradation degrees. Error degradation and time degradation The preset weighting coefficients; Spatial positioning accuracy index The formula is: ; The spindle performance index is obtained by acquiring the spindle vibration signal of the equipment and performing a fast Fourier transform on the spindle vibration signal to obtain the vibration spectrum. From the vibration spectrum, characteristic frequencies corresponding to the health status of key bearing components are identified and extracted, specifically including the characteristic frequencies of the bearing outer ring, inner ring, and rolling elements, and their corresponding energy amplitudes are obtained. The energy amplitudes of the outer ring, inner ring, and rolling elements are normalized according to their respective preset threshold values ​​to generate independent, dimensionless sub-items of vibration degradation. The sub-items of vibration degradation are linearly combined according to preset weighting coefficients that sum to 1 to generate a comprehensive vibration degradation degree that fully reflects the overall health status of the bearing. The spindle performance index is measured with a maximum score of 100, and its final score is obtained by subtracting a loss factor proportional to the magnitude of the comprehensive vibration degradation degree from the maximum score.

[0035] First, the outer ring degradation degree is calculated. Inner ring deterioration and rolling element deterioration :

[0036] in, This represents the energy amplitude corresponding to the characteristic frequency of the bearing outer ring. The preset outer ring energy amplitude threshold;

[0037] in, This represents the energy amplitude corresponding to the characteristic frequency of the bearing's inner ring. The preset threshold value for the inner circle energy amplitude;

[0038] in, This represents the energy amplitude corresponding to the characteristic frequency of the rolling element. This is a preset threshold value for the rolling element energy amplitude; Degradation of the outer ring Inner ring deterioration and rolling element deterioration After normalization, the overall vibration degradation degree is calculated using the following formula. :

[0039] in, , and The outer ring degradation Inner ring deterioration and rolling element deterioration The preset weighting coefficients; Spindle performance index The formula is: ; The tool stability index is obtained by using the tool clamping or releasing command as the starting trigger signal for data acquisition when executing the tool change command, and the corresponding travel limit switch being triggered or the pressure reaching a stable state as the termination signal, thereby capturing a first future time window. Within the first future time window, the output value of the pressure sensor installed in the hydraulic or pneumatic actuation circuit of the clamping or releasing mechanism is acquired at high frequency to form a dynamic pressure curve that can characterize the entire process of each clamping or releasing action. The dynamic pressure curve is then time-aligned with the reference pressure representing the healthy state pre-stored in the system, and the cumulative waveform distance between the two is calculated. At the same time, the deviation value between the actual duration of this action and the reference time is calculated to form the time deviation. The cumulative distance and time deviation of the waveform are normalized according to their preset thresholds to generate distance degradation and time degradation. The distance degradation and time degradation are weighted and combined to form a comprehensive offset metric. The tool stability index is measured with 100 points as the full score, and its final score is obtained by deducting the loss component proportional to the size of the comprehensive offset metric from the full score. First, the distance degradation degree is calculated. and motion time degradation :

[0040] in, The cumulative distance between the dynamic pressure curve and the reference pressure curve. The preset waveform cumulative distance threshold;

[0041] in, This represents the deviation between the actual duration of this action and the reference time. The preset time deviation threshold; Distance degradation and motion time degradation After normalization, the overall offset metric is calculated using the following formula. :

[0042] in, and Distance degradation and motion time degradation The preset weighting coefficients; The formula for the tool stability index is: : ; The thermal stability index is obtained by simultaneously collecting the temperature change time series of at least one key heat source of the equipment and the thermally induced displacement time series of key reference points of the machine tool. Through correlation analysis, the thermal sensitivity coefficient K is calculated. The thermal sensitivity coefficient K quantifies the displacement caused by a unit temperature change, representing the thermal response characteristics of the equipment. The thermal sensitivity coefficients in multiple working cycles are extracted to obtain a set of sensitivity coefficients, and the standard deviation of the set of sensitivity coefficients is calculated. The standard deviation is normalized according to its preset performance threshold to generate a dimensionless thermal instability. The thermal stability index is measured with a full score of 100. Its final score is obtained by deducting a loss factor proportional to the magnitude of the thermal instability from the full score.

[0043] Calculate thermal instability :

[0044] in, The standard deviation of the thermal sensitivity coefficient K set over multiple operating cycles; The preset standard deviation performance threshold; Thermal stability index The calculation formula is: Please refer to Table 1 below for details: Table 1: Sample of Spatial Positioning Accuracy Index

[0045] The sample of spindle performance indices is shown in Table 2: Table 2: Sample of Spindle Performance Indices

[0046] The sample of tool stability index is shown in Table 3: Table 3: Sample of Tool Stability Index

[0047] The samples of thermal stability indices are shown in Table 4: Table 4: Sample of Thermal Stability Index

[0048] It should be noted that in this technical solution, the calculation of the degradation degree of each performance parameter relies on a preset performance benchmark threshold, which represents the upper limit of the allowable fluctuation of the key performance indicators when the equipment is running within an acceptable performance range, and constitutes the watershed between "normal operation loss" and "significant performance degradation". The threshold is determined using a calibration logic based on multi-sample health status statistics. The specific steps are as follows: First, a baseline sample set is constructed. During the mass production or initial deployment phase, a batch (e.g., M units, M≥10) of equipment that has passed factory inspection and is confirmed to be in brand new or optimal health condition is selected. Next, each piece of equipment in the baseline sample set is run under standard operating conditions to collect its raw performance data, forming multiple independent raw datasets. Subsequently, statistical analysis is performed on each raw dataset (e.g., the set of maximum overshoot values ​​for all equipment), calculating the arithmetic mean and standard deviation of all values ​​in the dataset. Finally, a safety margin coefficient f (between 3 and 6) is introduced based on the performance monitoring sensitivity requirements. The fluctuation tolerance value is obtained by multiplying the standard deviation by this coefficient f, and then added to the arithmetic mean to determine the performance baseline threshold. The technical rationale for choosing this statistical method in this scheme is that, according to the "3σ rule" or the more stringent "6σ rule" of the normal distribution, it can be ensured that in a healthy group of devices, the vast majority (e.g., 99.7% or higher) of performance parameter fluctuations will fall within the range of μ±fσ. This provides a reliable statistical basis for distinguishing between normal fluctuations and abnormal degradation, avoiding the uncertainty of subjective settings. For example, for the maximum overshoot performance threshold, the calibration method for the preferred value of 15μm in Table 1 is as follows: by testing M healthy devices, the average value μ of their maximum overshoot dataset is obtained. ds The value is 2.5 μm, and the standard deviation σ is 2.5 μm. ds The value is 4.0 μm. In this embodiment, to ensure sufficient performance redundancy, a safety margin factor f of 3.125 is selected. The calculated performance baseline threshold T is... ds For: T ds =2.5+3.125×4.0=2.5+12.5=15μm; the other thresholds and weight calibration methods are based on the same principle and will not be elaborated here.

[0049] In this embodiment, the present invention constructs a digital twin skill model based on an Asset Management Shell (AAS), abstracting the equipment into four core skill dimensions directly related to processing quality: spatial positioning accuracy, spindle performance, tool stability, and thermal stability. The calculation of each index has a clear physical meaning and engineering basis. For example, the spatial positioning accuracy index directly originates from the dynamic error between servo commands and the actual position, while the spindle performance index is derived from the bearing characteristic frequencies in the vibration spectrum. This mechanism-based refined modeling ensures that the prediction results are no longer abstract "health scores," but rather performance indicators that can be traced back to their origins and clearly explained, providing a solid scientific foundation for subsequent maintenance decisions.

[0050] Secondly, it innovatively uses "failure momentum" as an early warning trigger mechanism, upgrading the early warning mode from "static threshold" to "dynamic trend," significantly enhancing the foresight and intelligence of the early warning. Traditional predictive maintenance triggers an alarm when a performance indicator falls below a fixed threshold, at which point performance degradation may already be nearing a critical point. This invention utilizes an LSTM neural network to predict the future degradation curves of various skill indices and innovatively defines "failure momentum" by calculating the slope of the trend line within a future time window. This momentum reflects the "acceleration" of performance degradation, triggering an early warning in the early stages when performance indicators are still within a safe range but the rate of degradation is abnormally fast. This keen capture of deterioration trends facilitates earlier maintenance windows, gaining valuable time for production scheduling and resource preparation, achieving true "prevention is better than cure."

[0051] Furthermore, it provides in-depth insights into the performance of critical subsystems of the equipment, enabling precise identification of hidden, non-traditional failure modes. The quantitative methods for each skill index in this invention delve into the minute details of equipment operation. For example, the "Tool Stability Index" does not monitor the machining process, but cleverly assesses the health of the tool clamping system by analyzing the deviation of the dynamic pressure curve of the clamping / releasing mechanism from a benchmark during tool changing. This is crucial for preventing micro-vibrations and machining quality degradation caused by insufficient clamping force. Similarly, the "Thermal Stability Index" assesses the consistency of the equipment's thermal behavior by evaluating the dispersion (standard deviation) of the thermal sensitivity coefficient, effectively identifying machining accuracy instability caused by decreased cooling system efficiency or structural stress changes. These innovative quantitative methods can capture early, hidden problems that are difficult to detect with traditional vibration or temperature monitoring, greatly expanding the coverage and depth of predictive maintenance.

[0052] Example 3 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figures 1 to 3 Specifically, the maintenance simulation analysis module includes a demand matching unit; The demand matching unit is used to automatically perform demand matching in response to the first warning command; When making equipment allocation decisions for production scheduling, a demand matching process is executed, and a lower limit threshold for failure momentum is preset. When the failure momentum of the i-th equipment in each dimension is lower than the lower limit threshold for failure momentum at future time t, it means that the equipment is qualified in the current skill dimension. The difference between the two is calculated to obtain the health margin. Based on the corresponding process of each skill dimension, the health margins are sorted from high to low to obtain the first sorting queue. The system continuously monitors the failure momentum of all devices. When the failure momentum of any device in any skill dimension exceeds the preset upper limit threshold, a first warning instruction is triggered. The first warning instruction includes maintenance data related to the device. The system calculates the excess value of the failure momentum exceeding the upper limit threshold and defines the excess value as the maintenance urgency index of the current device. Based on the maintenance urgency index, all devices that have triggered warnings are sorted from high to low to form a second sorting queue. According to the second sorting queue, the initial priority of maintenance tasks is obtained. The initial priority of maintenance tasks is positively correlated with the maintenance urgency index, that is, the higher the maintenance urgency index, the higher the initial priority is set, and the higher the resource allocation weight is enjoyed when matching with maintenance resource data. Maintenance resource data includes: available work schedules for each member of the maintenance team, real-time inventory and estimated delivery time of required spare parts, and non-production windows in the production plan where maintenance tasks can be inserted.

[0053] When the second sorting queue is not empty, the first maintenance planning process is activated according to the candidate devices in the second sorting queue, specifically as follows: Based on the task arrangement of the current candidate equipment in the production schedule, the task dependency of the current equipment is identified. The task dependency is determined based on the production task priority in the production schedule. The task dependency is obtained by adding the scores of the directly downstream tasks related to the equipment. If the candidate equipment is assigned a high-priority production task, then that equipment will be temporarily skipped, and maintenance planning will be prioritized for the next equipment in the second sorting queue that is not assigned a high-priority task.

[0054] The maintenance simulation analysis module also includes a simulation scheduling scheme unit and a comprehensive cost analysis unit. The simulation scheduling scheme unit is used to filter the equipment to be planned based on the first sorting queue, identify and generate multiple potential maintenance execution time windows based on the task arrangement of the equipment in the production schedule plan, and perform an independent scheduling recalculation simulation for each potential maintenance execution time window to generate a corresponding simulation scheduling scheme. Multiple potential maintenance execution time windows are used to generate multiple simulated scheduling schemes, forming the first potential maintenance execution time window column; The comprehensive cost analysis unit is used to analyze the corresponding simulated scheduling schemes in the first potential maintenance execution time window column to quantify the ripple effect caused by the insertion of maintenance tasks and construct the comprehensive cost value of the j-th corresponding simulated scheduling scheme for the i-th device.

[0055] Preferably, the method for obtaining the comprehensive cost value of the j-th corresponding simulated scheduling scheme for the i-th device is as follows: To calculate the cost of delayed delivery, identify all orders with delivery time delays in the simulated scheduling scheme, calculate the delay duration of each order, multiply it by a dynamic delay penalty coefficient that is positively correlated with the current order priority or customer level, and finally sum the calculation results of all orders to obtain the cost of delayed delivery. Calculate the cost of lost capacity by calculating the equipment downtime caused by the insertion of maintenance tasks, and multiply it by the unit time capacity value of the equipment. Calculate the indirect rescheduling cost, and calculate the additional material transfer time, new equipment process preparation time, and processing time increment caused by the efficiency difference of the alternative equipment when the affected process is moved to other alternative equipment to avoid delays. Convert these times into costs. To calculate supply chain disruption costs, analyze the changes in the completion time of all processes in the simulated scheduling scheme, quantify the advance or delay of the input material supply time of downstream processes, and calculate the cost by multiplying the amount of time drift by the supply chain stability penalty coefficient. Downstream processes include assembly or testing processes. The comprehensive cost value is obtained by directly adding the costs of delayed delivery, capacity loss, indirect rescheduling, and supply chain disruption.

[0056] The following is an example of calculating delayed delivery costs: A 2-hour maintenance window was planned for equipment CNC-01, which affected Order A (VIP customer, high priority) and Order B (regular user) being executed on it. The simulation plan decided to migrate the operation of Order B to the backup equipment CNC-02 for execution. Calculate the delay duration: Original planned completion time: 16:00; New plan completion time: 18:00; The delay time is calculated as 18:00 - 16:00 = 2 hours; Result: The delay time for order A is 2 hours; As VIP customers, the business impact of order delays is greater than that of regular orders. Therefore, a higher unit time is set, specifically, the delay penalty coefficient for order A is 500 cost units / hour. Calculate the delay cost for order A: Delay cost = Delay duration × Delay penalty coefficient; the result is 1000 cost units. Calculate the delay cost of order B. Order B was moved to the less efficient backup CNC-02 for execution, and its final completion time was delayed by 1 hour compared to the original plan. Determine the delay penalty coefficient. As a regular user order, its delay sensitivity is lower than that of a VIP order, therefore the penalty value per unit time is set relatively low. The delay penalty coefficient for order B is 150 cost units / hour. Calculate the delay cost of order B: Delay cost = Delay duration × Delay penalty coefficient; the result is 150 cost units. The final delayed delivery cost is obtained by summing up the deferred costs of all affected orders. In this example, 1000 + 150 = 1150 cost units.

[0057] The following is an example of calculating the cost of capacity loss: Maintenance task: Perform planned maintenance on equipment CNC-01; Downtime: 2 hours; Determine the total downtime of equipment due to maintenance. According to the maintenance plan, the downtime of equipment CNC-01 is 2 hours. Determine the unit time capacity value of the equipment, for example, the capacity value of equipment CNC-01 is 150 cost units / hour. The specific calculation of capacity loss cost is as follows: Capacity loss cost = total downtime × capacity value per unit time. Substituting this into the calculation, we get 2 hours × 150 cost units / hour, resulting in 300 cost units.

[0058] The following is an example of calculating the cost of indirect rescheduling: The process for order B, originally planned to be executed on CNC-01, will be transferred to the backup equipment CNC-02. First, calculate the material handling cost, which is the logistics expense incurred in moving the semi-finished products and raw materials required for Order B from the CNC-01 work area to the CNC-02 work area. The material handling cost is 30 cost units. Then, calculate the production setup cost on the new equipment. This includes the labor and resource costs incurred on the CNC-02 for machine tool debugging, tool / fixture replacement, program calling, and other preparatory activities for the Order B process. The production setup cost is 50 cost units. Calculate the additional time cost caused by differences in equipment efficiency: Calculate the additional time consumed: Originally planned machining time on CNC-01: 1.0 hour; Current machining time on CNC-02: 1.5 hours (example, CNC-02 is less efficient); Calculation: 1.5 hours - 1.0 hour = 0.5 hours; Result: The additional time taken was 0.5 hours.

[0059] Calculate the additional time cost: This extra time consumes CNC-02's capacity, so the loss should be calculated using the capacity value of CNC-02.

[0060] The unit time production value of CNC-02: 120 cost units / hour.

[0061] Calculation: 0.5 hours × 120 cost units / hour = 60 cost units.

[0062] Finally, calculate the indirect rescheduling cost: material transfer cost + production preparation cost + additional time cost, and substitute into the calculation: 30 cost units + 50 cost units + 60 cost units = 140 cost units.

[0063] The following is an example of calculating supply chain disruption costs: Supply chain disruption costs quantify the impact of changes in production plans on downstream processes. Whether delivering finished products to upstream processes is done ahead of schedule or delayed, it disrupts the production rhythm of downstream processes and generates coordination costs.

[0064] The supply chain relationship is as follows: The process completed by order A on CNC-01 produces materials that are necessary for the downstream "assembly process"; First, calculate the time drift of material supply to downstream processes: Original scheduled supply time: 14:00; New supply schedule: 16:00; Calculation: |16:00-14:00| = 2 hours; Result: The time drift is 2 hours. The absolute value is used here because even if the delivery is 2 hours early, downstream processes will still need to adjust their plans to cope, which is also considered a disturbance.

[0065] Next, determine the disturbance penalty coefficient for downstream processes. This coefficient represents the loss (such as worker waiting, production line rearrangement, etc.) caused by a one-hour deviation from the production plan of the downstream "assembly process".

[0066] Setting value: Disturbance penalty coefficient is 80 cost units / hour.

[0067] The specific calculation of supply chain disruption costs is as follows: time drift × disruption penalty coefficient = 2 hours × 80 cost units / hour = 160 cost units; The maintenance simulation analysis module also includes a candidate window pruning unit and a cooperative scheduling instruction generation unit; The candidate window pruning unit is used to evaluate each simulated scheduling scheme to quantify the energy consumption effect caused by the insertion of maintenance tasks and construct the total energy consumption value of the j-th corresponding simulated scheduling scheme for the i-th device. The total energy consumption value of the i-th device in the j-th corresponding simulated scheduling scheme is obtained by multiplying the running time of each device in the corresponding simulated scheduling scheme by the power of the corresponding time period. Set a carbon footprint consumption threshold. When the total energy consumption of the j-th corresponding simulated scheduling scheme of the i-th device exceeds the carbon footprint consumption threshold, the current scheduling scheme will be removed from the potential maintenance execution time window to form a second potential maintenance execution time window column. The simulated scheduling schemes corresponding to the second potential maintenance execution time window column are traversed, and the potential maintenance execution time window with the smallest comprehensive cost value is selected as the optimal solution and the optimal maintenance intervention time window, and the first collaborative scheduling instruction is generated.

[0068] In this embodiment, existing technologies process maintenance requests using a "first-come, first-served" or fixed rules. This invention innovatively establishes a two-way prioritization mechanism: on the one hand, a first prioritization queue formed by "health margin" can proactively match newly entering processes with equipment in optimal health status, achieving opportunistic prevention; on the other hand, a second prioritization queue formed by "maintenance urgency index" can dynamically and quantitatively prioritize equipment that has triggered warnings based on the degree to which failure momentum exceeds a threshold. More importantly, the system can identify the "task dependency" of the equipment to be maintained and intelligently postpone maintenance planning for equipment currently performing high-priority production tasks. This dynamic prioritization mechanism, which considers both equipment health trends and the value of production tasks, ensures that limited maintenance resources are always allocated to the most impactful and urgent aspects of overall production, achieving intelligent and optimized resource allocation.

[0069] Traditional maintenance decisions often involve vague and one-sided assessments of downtime impact. This invention, through a comprehensive cost analysis unit, decomposes and quantifies the complex chain reaction (ripple effect) triggered by inserting a maintenance task into four dimensions: delayed delivery costs (external customer impact), capacity loss costs (internal asset efficiency), indirect rescheduling costs (operational flexibility costs), and supply chain disruption costs (process stability impact). This model unifies different types of impacts onto a universal metric of "comprehensive cost value," allowing for a direct and objective comparison of the merits of different maintenance window options. This fundamentally changes the traditional model that relies on manual experience to weigh pros and cons, providing a solid and reliable quantitative basis for finding the maintenance timing with the lowest overall cost.

[0070] Existing technologies, when conducting production scheduling and maintenance planning, almost entirely focus on time and cost, neglecting energy consumption and environmental impact. This invention, through a candidate window pruning unit, first assesses the energy consumption of each simulated scheduling scheme before cost optimization, and establishes a "carbon footprint consumption threshold" as a hard filter condition to directly eliminate high-energy-consuming schemes. This two-stage optimization strategy of "environmental protection first, economy second" deeply embeds the concept of green manufacturing into the core decision-making logic, ensuring that the final optimal solution is not only the one with the lowest economic cost but also meets the requirements of sustainable development for the enterprise. This allows the system decision-making to consider long-term social and environmental responsibility while pursuing short-term economic benefits, greatly enhancing the strategic height and comprehensive value of the decision.

[0071] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0072] The above formulas are all derived from software simulation using a large amount of data and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the protection scope of the present invention.

Claims

1. An intelligent predictive maintenance system for equipment integrating production scheduling, characterized in that: include: The equipment prediction module is used to establish a structured data model for each piece of equipment in the production system, which is directly related to the processing quality of the skill dimensions. It acquires the operating parameter data of the equipment in real time and converts the operating parameter data into quantitative index values ​​for each skill dimension in the skill profile. The quantitative index values ​​include, but are not limited to, spatial positioning accuracy index, spindle performance index, tool stability index, and thermal stability index. The real-time generated quantitative index values ​​are recorded in chronological order to generate a historical quantitative index value sequence for each piece of equipment. A time series prediction model is used to generate a future decline trend curve for each skill dimension and generate the failure momentum of the i-th device in each dimension at future time t. A failure momentum upper limit threshold is preset. When the failure momentum of each skill dimension exceeds the failure momentum upper limit threshold, the first warning instruction is triggered. The maintenance simulation analysis module is used to execute a demand matching process when making equipment allocation decisions in production scheduling after receiving the first early warning instruction. It evaluates and obtains the first sorting queue and the second sorting queue. It filters in the first sorting queue and identifies and generates multiple potential maintenance execution time windows based on the task arrangement of the equipment in the production scheduling plan, forming the first potential maintenance execution time window column. The simulation scheduling schemes corresponding to the first potential maintenance execution time window column are analyzed to quantify the ripple effect caused by the insertion of maintenance tasks and to construct the comprehensive cost value and total energy consumption value. A carbon footprint consumption threshold is preset, and simulated scheduling schemes with total energy consumption exceeding the carbon footprint consumption threshold are eliminated. The potential maintenance execution time window with the lowest comprehensive cost value is selected as the optimal solution, and the first collaborative scheduling instruction is generated.

2. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The equipment prediction module includes an equipment skill profile construction unit and a failure prediction unit; The skill profile building unit is used to build a skill profile for each device, create a digital twin skill model for each device, acquire the operating parameter data of the device in real time, and perform training and verification to build the spatial positioning accuracy index, spindle performance index, tool stability index and thermal stability index of each device. The failure prediction unit is used to generate a sequence of historical quantitative index values ​​based on the spatial positioning accuracy index, spindle performance index, tool stability index and thermal stability index of each device. It uses LSTM neural network technology to build a time series prediction model, trains the historical quantitative index value sequence, draws the future decline trend curve for each skill dimension, and generates the failure momentum of each dimension of the i-th device at future time t. The decline trend curve consists of a set of multi-dimensional core performance index prediction values ​​for future time t, covering spatial positioning accuracy, spindle performance, tool stability, and thermal stability. The method for obtaining the failure momentum of each skill dimension in the i-th device at future time t is as follows: Select a future time window, from the current time tn to the future time tn+k after k time steps, to obtain the first future time window. For each skill dimension, use linear regression to fit the predicted index value sequence within the first future time window, and extract the slope of the fitted line as the failure momentum of that skill dimension. The length of the first future time window is set based on the historical mean time between failures (MTBF) of the device; A preset upper limit threshold for failure momentum is set. When the failure momentum of each skill dimension exceeds the upper limit threshold, the first warning instruction is triggered.

3. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The spatial positioning accuracy index is obtained by acquiring the servo motor command position signal of the device and the actual position signal fed back by the grating ruler or encoder, calculating the dynamic position error between the two, obtaining an error sequence, extracting the maximum overshoot, root mean square error, and settling time that characterize the positioning performance degradation from the error sequence, and normalizing them according to the preset thresholds corresponding to the maximum overshoot, root mean square error, and settling time to generate independent, dimensionless sub-items of positioning degradation. The sub-items of positioning degradation are then linearly combined using preset weighting coefficients to generate a composite positioning degradation. The composite positioning degradation is negatively correlated with the spatial positioning accuracy index, which is measured with a maximum score of 100. The final score is obtained by subtracting a loss factor proportional to the magnitude of the composite positioning degradation from the maximum score.

4. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The spindle performance index is obtained by acquiring the spindle vibration signal of the equipment and performing a fast Fourier transform on the spindle vibration signal to obtain the vibration spectrum. From the vibration spectrum, characteristic frequencies corresponding to the health status of key bearing components are identified and extracted, specifically including the characteristic frequencies of the bearing outer ring, inner ring, and rolling elements, and their corresponding energy amplitudes are obtained. The energy amplitudes of the outer ring, inner ring, and rolling elements are normalized according to their preset threshold values ​​to generate independent, dimensionless sub-items of vibration degradation. The sub-items of vibration degradation are linearly combined according to preset weighting coefficients that sum to 1 to generate a comprehensive vibration degradation degree that fully reflects the overall health status of the bearing. The spindle performance index is measured with a maximum score of 100, and its final score is obtained by deducting a weight proportional to the magnitude of the comprehensive vibration degradation degree from the maximum score.

5. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The tool stability index is obtained by using the command to control tool clamping or releasing as the starting trigger signal for data acquisition when executing the tool change command, and the signal that the corresponding travel limit switch is triggered or the pressure reaches a stable state as the termination signal, thereby capturing a first future time window. Within the first future time window, the output value of the pressure sensor installed in the hydraulic or pneumatic execution circuit of the clamping or releasing mechanism is acquired at high frequency to form a dynamic pressure curve that can characterize the entire process of each clamping or releasing action. The dynamic pressure curve is time-aligned with the reference pressure curve representing the healthy state that is pre-stored in the system, and the cumulative waveform distance between the two is calculated. At the same time, the deviation value between the actual duration of this action and the reference time is calculated to form the time deviation. The cumulative distance and time deviation of the waveform are normalized according to their preset threshold values ​​to generate distance degradation and time degradation. The distance degradation and time degradation are then weighted and combined to form a comprehensive offset metric. The tool stability index is measured out of 100 points. Its final score is obtained by subtracting a loss factor that is proportional to the magnitude of the overall offset metric from the full score. The thermal stability index is obtained by simultaneously collecting the temperature change time series of at least one key heat source of the equipment and the thermal displacement time series of key reference points of the machine tool. Through correlation analysis, the thermal sensitivity coefficient K is calculated. The thermal sensitivity coefficient K quantifies the displacement caused by a unit temperature change and represents the thermal response characteristics of the equipment. The thermal sensitivity coefficients in multiple working cycles are extracted to obtain a set of sensitivity coefficients. The standard deviation of the set of sensitivity coefficients is calculated and normalized according to its preset performance threshold to generate a dimensionless thermal instability. The thermal stability index is measured with a full score of 100. Its final score is obtained by deducting a loss factor proportional to the magnitude of thermal instability from the full score.

6. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The maintenance simulation analysis module includes a demand matching unit; The demand matching unit is used to automatically perform demand matching in response to the first warning instruction; When making equipment allocation decisions for production scheduling, a demand matching process is executed, and a lower limit threshold for failure momentum is preset. When the failure momentum of the i-th equipment in each dimension is lower than the lower limit threshold for failure momentum at future time t, it means that the equipment's current skill dimension is qualified. The difference between the two is calculated to obtain the health margin. Based on the corresponding process for each skill dimension, the health margins are sorted from high to low to obtain the first sorted queue; Continuously monitor the failure momentum of all devices. When the failure momentum of any skill dimension of any device exceeds the preset failure momentum upper limit threshold, trigger the first warning instruction. The first warning instruction includes the maintenance data related to the device. Calculate the excess difference when the failure momentum exceeds the upper limit threshold of the failure momentum, and define the excess difference as the maintenance urgency index of the current equipment; based on the maintenance urgency index, sort all equipment that triggers the warning from high to low to form a second sorting queue; according to the second sorting queue, obtain the initial priority of the maintenance task; the initial priority of the maintenance task is positively correlated with the maintenance urgency index, that is, the larger the maintenance urgency index, the higher the initial priority is set, and the higher the resource allocation weight is enjoyed when matching with maintenance resource data; The maintenance resource data includes: available work schedules for each member of the maintenance team, real-time inventory and estimated delivery time of required spare parts, and non-production windows in the production plan where maintenance tasks can be inserted.

7. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 6, characterized in that, When the second sorting queue is not empty, the first maintenance planning process is activated according to the candidate devices in the second sorting queue, specifically as follows: Based on the task arrangement of the current candidate equipment in the production schedule, the task dependency of the current equipment is identified. The task dependency is determined based on the production task priority in the production schedule. The task dependency is obtained by adding the scores of the directly downstream tasks related to the equipment. If the candidate equipment is assigned a high-priority production task, then that equipment will be temporarily skipped, and maintenance planning will be prioritized for the next equipment in the second sorting queue that is not assigned a high-priority task.

8. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The maintenance simulation analysis module also includes a simulation scheduling scheme unit and a comprehensive cost analysis unit. The simulation scheduling scheme unit is used to filter the equipment to be planned based on the first sorting queue, identify and generate multiple potential maintenance execution time windows according to their task arrangements in the production scheduling plan, and perform an independent scheduling recalculation simulation for each potential maintenance execution time window to generate a corresponding simulation scheduling scheme. Multiple potential maintenance execution time windows are generated, resulting in multiple simulated scheduling schemes, forming the first potential maintenance execution time window column; The comprehensive cost analysis unit is used to analyze the corresponding simulated scheduling schemes in the first potential maintenance execution time window column to quantify the ripple effect caused by the insertion of maintenance tasks and construct the comprehensive cost value of the j-th corresponding simulated scheduling scheme for the i-th device.

9. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 8, characterized in that, The specific method for obtaining the comprehensive cost value of the j-th corresponding simulated scheduling scheme for the i-th device is as follows: To calculate the cost of delayed delivery, identify all orders with delivery time delays in the simulated scheduling scheme, calculate the delay duration of each order, multiply it by a dynamic delay penalty coefficient that is positively correlated with the current order priority or customer level, and finally sum the calculation results of all orders to obtain the cost of delayed delivery. Calculate the cost of lost capacity by calculating the equipment downtime caused by the insertion of maintenance tasks, and multiply it by the unit time capacity value of the equipment. Calculate the indirect rescheduling cost, and calculate the additional material transfer time, new equipment process preparation time, and processing time increment caused by the efficiency difference of the alternative equipment when the affected process is moved to other alternative equipment to avoid delays. Convert these times into costs. To calculate supply chain disruption costs, analyze the changes in the completion time of all processes in the simulated scheduling scheme, quantify the advance or delay of the input material supply time of downstream processes, and calculate the cost by multiplying the amount of time drift by the supply chain stability penalty coefficient. Downstream processes include assembly or testing processes. The comprehensive cost value is obtained by directly adding the costs of delayed delivery, capacity loss, indirect rescheduling, and supply chain disruption.

10. The intelligent predictive maintenance system for equipment integrating production scheduling according to claim 1, characterized in that, The maintenance simulation analysis module also includes a candidate window pruning unit and a collaborative scheduling instruction generation unit; The candidate window pruning unit is used to evaluate each simulated scheduling scheme to quantify the energy consumption effect caused by the insertion of maintenance tasks and construct the total energy consumption value of the j-th corresponding simulated scheduling scheme for the i-th device. The total energy consumption value of the i-th device in the j-th corresponding simulated scheduling scheme is obtained by multiplying the running time of each device in the corresponding simulated scheduling scheme by the power of the corresponding time period. Set a carbon footprint consumption threshold. When the total energy consumption of the j-th corresponding simulated scheduling scheme of the i-th device exceeds the carbon footprint consumption threshold, the current scheduling scheme will be removed from the potential maintenance execution time window to form a second potential maintenance execution time window column. The simulated scheduling schemes corresponding to the second potential maintenance execution time window column are traversed, and the potential maintenance execution time window with the smallest comprehensive cost value is selected as the optimal solution and the optimal maintenance intervention time window, and the first collaborative scheduling instruction is generated.

Images