Intelligent manufacturing-oriented device health management and fault prediction and early warning method

Abstract

This invention discloses a method for equipment health management and fault prediction and early warning in intelligent manufacturing, aiming to solve the problems of disconnect between fault prediction and production scheduling, and the lack of business relevance to early warning information, which makes closed-loop execution difficult. The method includes: real-time collection of multi-dimensional equipment operating data and extraction of feature vectors; identification of fault types and remaining safe operating time through a model; generation of a structured maintenance suggestion package containing fault level, maintenance skill tags, spare parts codes, and recommended window duration by associating with a maintenance knowledge base; pushing the suggestion package to the manufacturing execution system (MES) via a cross-system interface to trigger a work order; and the MES initiating a production plan re-optimization algorithm to implement task migration and reserve maintenance windows while meeting delivery constraints. This application constructs a closed-loop mechanism of "early warning-decision-scheduling," breaks down information silos, achieves flexible migration of production tasks and precise matching of maintenance resources, and significantly improves the response speed and operational reliability of the manufacturing system.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing technology, specifically, it relates to a method for equipment health management and fault prediction and early warning for intelligent manufacturing. Background Technology

[0002] With the deepening advancement of intelligent manufacturing in the global industrial sector, the integration and intelligence of industrial equipment are constantly increasing, posing higher-level challenges to health management and fault prediction and early warning throughout the entire equipment lifecycle. As a core component ensuring the stable operation of industrial production, equipment health management technology aims to build a monitoring system that covers the entire process of equipment degradation by capturing state parameters in complex operating environments in real time, thereby improving the overall reliability and safety of the production system.

[0003] Among them, fault prediction and early warning technology for dynamic production environments focuses on using multi-dimensional sensor data to quantitatively assess the health status of key equipment components. This technology attempts to identify potential performance degradation trends before a fault occurs by deeply mining and extracting features from massive heterogeneous data, and formulates targeted maintenance strategies based on this. Its core objective is to achieve optimal allocation of production resources and maximize the release of equipment operating efficiency.

[0004] Existing technologies for equipment health management are often constrained by architectural barriers between systems, leading to a significant disconnect between fault prediction and actual production scheduling. Traditional monitoring methods primarily focus on signal recognition at the perception level, and their output warning information typically lacks key business elements such as maintenance skill tags, necessary spare parts material codes, and estimated maintenance durations, making it difficult to directly translate into a structured maintenance recommendation package that can be implemented. Simultaneously, the lack of an efficient interface linkage mechanism between the health management system and the manufacturing execution system prevents warning information from triggering dynamic restructuring of production plans in real time, hindering the flexible migration of production tasks and precise reservation of maintenance windows when dealing with sudden failures. Furthermore, the lack of a closed-loop collaborative logic integrating warning, decision-making, and scheduling frequently leads to conflicts between equipment maintenance activities and production delivery schedules, severely restricting the response speed and overall execution efficiency of the manufacturing system when handling high-priority faults. Summary of the Invention

[0005] The purpose of this invention is to provide a method for equipment health management and fault prediction and early warning in intelligent manufacturing, which mainly solves the problems of fault prediction being disconnected from production scheduling and the lack of business relevance to early warning information, making it difficult to implement in a closed loop.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for equipment health management and fault prediction and early warning in intelligent manufacturing includes the following steps:

[0008] S1 utilizes a sensor array deployed in key parts of the manufacturing equipment to collect and process real-time operating status parameters of the equipment, obtaining multi-dimensional raw data;

[0009] S2, acquire the time domain features, frequency domain features and spatial domain features of the multidimensional raw data, construct the equipment health status feature vector, and input the equipment health status feature vector into the preset fault prediction model;

[0010] S3 identifies potential faults such as bearing wear or spindle imbalance in manufacturing equipment and outputs the fault type and its corresponding remaining safe operating time.

[0011] S4. Based on the fault type and the remaining safe operating time of the equipment, a structured maintenance suggestion package is automatically calculated and generated by linking the maintenance knowledge base and the spare parts management database.

[0012] S5, the structured maintenance suggestion package is pushed to the maintenance management module of the manufacturing execution system in real time. The manufacturing execution system parses the data items in the structured maintenance suggestion package and triggers the automated maintenance work order creation process and material reservation instructions.

[0013] S6, start the production plan re-optimization algorithm, and under the premise of meeting the delivery date constraints and process constraints, migrate the subsequent production tasks of the affected equipment to the standby production unit, and reserve a certain maintenance execution window for the affected equipment to ensure that maintenance personnel and maintenance materials have arrived at the designated equipment location when the maintenance execution window is opened.

[0014] Further, in step S1, the sensor array includes multiple triaxial vibration acceleration sensors, multiple high-precision infrared temperature sensors, and at least one Hall current sensor; wherein, the frequency response of the triaxial vibration acceleration sensor is within a preset frequency range, and the sampling frequency is set to a preset sampling frequency; the measurement accuracy of the high-precision infrared temperature sensor is better than a preset accuracy threshold; the Hall current sensor is connected in series at the output end of the spindle frequency converter to monitor the current fluctuation caused by changes in cutting load in real time.

[0015] Furthermore, in step S3, the remaining safe operating time of the device is calculated based on the Weibull distribution function, combined with the real-time extracted feature vector offset, to calculate the current reliability index of the device. When the reliability index is lower than the preset reliability threshold, it is determined to be in a warning state, and the specific remaining lifespan value is output.

[0016] Furthermore, in step S4, the structured maintenance recommendation package includes the severity level of the fault, the skill tags required for maintenance, the material codes of necessary spare parts, and the recommended maintenance window duration.

[0017] Furthermore, the method for determining the recommended maintenance window duration is as follows: extract the historical average maintenance time for similar faults from the maintenance knowledge base, combine it with the current physical space layout of the equipment and the estimated time for spare parts to arrive, calculate the total time required to complete the maintenance, and add a preset proportion of redundant buffer time; the preset level in the fault severity level represents the highest priority, which requires downtime processing within a preset time period.

[0018] Furthermore, in step S6, the production planning re-optimization algorithm adopts an improved genetic algorithm or ant colony algorithm. Its objective function is set to minimize the total order delay time and minimize the equipment switching cost. The constraints include the load rate of standby equipment not exceeding a preset load rate threshold, real-time availability of material supply, and consistency of product process paths.

[0019] Further, in step S6, the method for determining the migration of the production task is as follows: identify the process currently being executed by the affected equipment and determine the completion progress of the process; if the completion progress is less than a first preset progress threshold, the current process is immediately interrupted and migration is performed; if the completion progress is greater than a second preset progress threshold, the migration is performed after the current process is completed, and the material flow diagram in the manufacturing execution system is automatically updated.

[0020] Furthermore, before launching the production planning re-optimization algorithm, a simulation verification step based on digital twins is also included: multiple candidate schemes generated by the algorithm are virtually run in a digital twin model to evaluate the impact of different scheduling strategies on output within a preset time period, and the optimal scheme is selected to perform task migration.

[0021] Furthermore, this method also includes establishing a closed-loop feedback mechanism for equipment health status: after maintenance is completed, the equipment operation data is re-collected through the sensor array to verify whether the fault has been eliminated; the maintenance process data, actual time consumption data and spare parts consumption data are fed back to the maintenance knowledge base, and the maintenance knowledge base is self-learned and dynamically updated using reinforcement learning algorithms.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) This invention transforms simple fault warning signals into structured maintenance suggestion packages containing fault level, maintenance skill tags, spare parts codes, and recommended window duration by linking the maintenance knowledge base and spare parts management database. This solves the problem that traditional warning information lacks business-related elements and cannot be directly implemented. Warning information can directly trigger maintenance resource scheduling without manual secondary sorting, shortening the preparation time for fault response and greatly improving the timeliness of fault handling.

[0024] (2) This invention constructs a standardized interface linkage mechanism between the health management system and the manufacturing execution system. Fault warning information can be pushed to the manufacturing execution system in real time to trigger the creation of automated work orders and material reservation. At the same time, through the production plan re-optimization algorithm, the production task is flexibly migrated and the maintenance window is accurately reserved under the constraints of delivery date and process. This fundamentally solves the pain point of the disconnect between fault prediction and production scheduling in traditional technology. It avoids production interruption caused by unplanned downtime and ensures that the order delivery cycle is not affected by equipment maintenance, thereby reducing the loss from unplanned downtime.

[0025] (3) This invention establishes a closed-loop operation and maintenance logic of “early warning-decision-scheduling-verification-feedback”. After the maintenance is completed, the maintenance effect is verified by collecting data from the sensors, and the actual maintenance process data and spare parts consumption data are fed back to the maintenance knowledge base. The knowledge base is updated by self-learning through reinforcement learning algorithm, so that the fault prediction accuracy, maintenance suggestion matching degree and window duration estimation accuracy continue to improve with the running time, and the overall efficiency of equipment operation and maintenance can be improved in the long term.

[0026] (4) This invention collects multi-dimensional operating data such as equipment vibration, temperature and current through multi-type sensor arrays, and combines time domain, frequency domain and spatial domain feature extraction with Weibull distribution model to accurately calculate the remaining safe operating time. At the same time, digital twin simulation verification is introduced in the production plan re-optimization process, and the candidate scheduling scheme is evaluated in virtual operation before execution, which effectively avoids the resource mismatch problem caused by traditional experience-based scheduling. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall scheme architecture of the present invention;

[0028] Figure 2 This is a schematic diagram of the core principle framework of the equipment health status feature extraction and fault prediction model in this invention;

[0029] Figure 3 This is a logical flowchart of the cross-system interface linkage and dynamic re-optimization of production plan in this invention. Detailed Implementation

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0031] like Figure 1 As shown, the equipment health management and fault prediction and early warning method for intelligent manufacturing disclosed in this invention establishes a complete hardware system covering the perception layer, edge layer, network layer, and application layer. The perception layer is deployed in key parts of each machining center, such as the machine tool spindle, feed axis screw, tool magazine, and hydraulic pump station. The sensor array specifically includes four triaxial vibration acceleration sensors mounted on the front and rear bearing seats of the spindle, used to capture weak vibration signals during the high-speed rotation of the spindle. Their frequency response range covers 10Hz to 20kHz, ensuring coverage of characteristic frequencies from low-frequency imbalance to high-frequency bearing defects; two non-contact high-precision infrared temperature sensors, respectively pointing to the spindle bearing and the drive motor housing, with a measurement accuracy of ±0.1℃, used to monitor temperature rise trends; and one through-hole Hall current sensor, connected in series at the output of the spindle frequency converter, used to monitor current fluctuations caused by changes in cutting load in real time.

[0032] The edge layer consists of industrial edge computing gateways deployed within the machine tool control cabinet. These gateways employ high-performance multi-core ARM processors and integrate a 16-bit multi-channel synchronous acquisition analog-to-digital converter. The sampling frequency is uniformly set to 51.2kHz to meet the Nyquist sampling law requirements for high-frequency vibration signal reproduction. The edge computing gateway communicates with the machine tool's CNC system via an RS485 interface to acquire real-time context information such as spindle speed, feed rate, and current program number. The network layer relies on a 5G industrial private network deployed within the workshop. Leveraging the high bandwidth and low latency of 5G, it transmits the processed feature data from the edge to the cloud server in the factory's central data center.

[0033] The specific execution flow of this system is as follows:

[0034] First, real-time operating status parameters of the equipment are collected and processed. The sensor array converts the collected analog electrical signals into digital sequences, and the edge computing gateway synchronously samples the raw data. Addressing the non-stationary nature of the vibration signal, the gateway's built-in digital signal processor performs a preprocessing procedure, using a wavelet packet decomposition algorithm to perform a 5-level decomposition of the signal, breaking it down into 32 independent frequency bands. The processor calculates the energy proportion of each frequency band and, combined with an empirical mode decomposition algorithm, decomposes the original signal into several intrinsic mode functions. By calculating the information entropy of each component, the system can identify the randomness and complexity variations in the signal. To achieve data dimensionality reduction while retaining core information, the system uses principal component analysis to transform the extracted 48-dimensional original features, retaining only the top 12 principal component features with a contribution rate greater than 95%, constructing a feature vector for the equipment's health status.

[0035] Subsequently, as Figure 2 As shown, the edge computing gateway inputs the feature vector into a pre-defined fault prediction model. This model employs a deep residual shrinking network structure, containing 18 residual modules and incorporating an attention mechanism. During inference, the model automatically identifies and eliminates background environmental noise interference, accurately identifying potential faults such as bearing fatigue pitting, ball screw wear, or spindle dynamic imbalance. The model outputs the specific fault type and calculates the equipment's reliability index using the Weibull distribution function. The reliability R(t) is calculated using the following formula:

[0036]

[0037] Where t is the cumulative operating time of the equipment. For scale parameters, The shape parameter is used. The system dynamically corrects the parameter by combining the real-time extracted feature vector offset. When the calculated reliability is lower than 0.6, the system determines that the equipment has entered an early warning state and outputs the specific remaining safe operating time in hours.

[0038] When the fault prediction model determines that the spindle bearing has a level 3 wear risk and the remaining safe operating time is 48 hours, the central decision-making platform automatically calls the maintenance knowledge base. The knowledge base uses a fuzzy matching algorithm to retrieve the standard operating procedure for replacing the spindle bearing for this machine model. Simultaneously, the system links with the spare parts management database to query the inventory status of P2-level precision bearings in real time. Finally, the system automatically generates a maintenance suggestion package in structured text object representation format, clearly indicating the fault severity as level 3 (medium to high), the required skill tag as "senior mechanic," the necessary spare parts material code as "BRG-2023-X102," and calculating a recommended maintenance window duration of 3.5 hours (including 15% operational redundancy) based on historical maintenance data.

[0039] The system then uses a RESTful application programming interface based on the secure hypertext transfer protocol to encapsulate the structured maintenance suggestion package in a JSON message and send it to the maintenance management module of the manufacturing execution system. OAuth 2.0 authentication is used during the interface call to ensure secure data transmission, and the end-to-end response latency is controlled within 100 milliseconds. Upon receiving the message, the manufacturing execution system automatically parses the data items and creates a pending work order on the maintenance dashboard. Simultaneously, it sends a spare parts reservation instruction to the materials management system to ensure that spare parts have been pre-sorted and stored in the temporary storage area when maintenance personnel collect materials.

[0040] Finally, dynamic re-optimization and collaborative scheduling of the production plan are implemented. For example... Figure 3As shown, after identifying a Level 3 fault warning and a 3.5-hour maintenance window requirement, the scheduling engine in the manufacturing execution system immediately initiates an improved genetic algorithm for production rescheduling optimization. The objective function of the algorithm is set as follows:

[0041]

[0042] in, The actual completion time of order i. For delivery date, As weight, To reduce equipment switchover and logistics relocation costs, the scheduling engine retrieves the affected equipment's production plan for the next 48 hours and finds that the equipment is currently performing a batch of high-precision piston pin finishing tasks. Since the task is 75% complete, according to the preset task migration logic, the system decides to allow the machine tool to continue completing the current process. Simultaneously, the system identifies another idle machine tool of the same specifications in the workshop as a backup production unit, automatically migrating the remaining 500 unstarted processing tasks to the backup machine tool and updating the path planning of the material handling robot. After the current process is completed, the system locks the affected equipment in the manufacturing execution system, reserves a specific maintenance execution window for it, and pushes precise start-up instructions to the maintenance worker's handheld terminal.

[0043] In this embodiment, through deep collaboration between sensor arrays, edge computing gateways, cloud decision-making platforms, and manufacturing execution systems, a closed-loop management system is achieved, encompassing physical perception, logical decision-making, and execution scheduling. The system can not only predict faults 48 hours in advance but also scientifically schedule maintenance plans without disrupting the overall production rhythm, reducing unplanned downtime losses by more than 85%.

[0044] Taking a certain automated precision machining production line as an example, the production line contains more than 120 key manufacturing equipment of various types, covering CNC machining, precision assembly, online testing, material transfer and other types of production equipment. It can be adapted to the large-scale flexible production of various products such as mechanical parts, general structural parts, and precision tooling.

[0045] In terms of system architecture, this embodiment adopts a centralized management platform architecture. The perception layer consists of over 800 sensors of various types distributed throughout the entire system. To handle the concurrent access of massive amounts of data, a distributed edge computing node cluster is deployed at the edge, with each node responsible for monitoring 5 to 8 adjacent devices. The edge nodes aggregate the processed feature vectors to the central server via a lightweight message queue transmission protocol. The central server adopts a distributed computing framework, supporting the input and real-time comparison of over 100,000 feature vectors per second.

[0046] When the main drive motor of the core processing equipment experiences abnormal current fluctuations, the system will simultaneously analyze the operating condition sensor data of adjacent process equipment. Through cross-equipment data correlation analysis, false alarms caused by external factors such as raw material size fluctuations and tooling clamping deviations can be eliminated, thereby improving the scientific nature of fault prediction.

[0047] The generated structured maintenance suggestion package adds two dimensions: "List of Tools Required for Maintenance" and "Safety Protection Level". For maintenance of high-voltage, high-speed rotating, or high-temperature parts, the system will automatically add mandatory safety instructions such as power-off tagging, wearing insulated gloves, and setting up safety guardrails to the suggestion package.

[0048] The interface linkage mechanism has added two-way confirmation logic. After receiving the suggestion package and creating a work order, the Manufacturing Execution System (MES) returns a work order serial number to the Health Management System. The Health Management System uses this number to track the execution progress of the work order and simultaneously displays the equipment's real-time health curve on the maintenance worker's mobile terminal to assist in fault location.

[0049] In this embodiment, the production planning re-optimization algorithm incorporates a simulation verification step based on digital twins. Before officially issuing the rescheduling instruction, multiple candidate solutions generated by the algorithm are virtually run in the digital twin model to evaluate the impact of different scheduling strategies on the output, work-in-process turnover, and equipment utilization rate in the subsequent 24 hours. Finally, the optimal solution is selected, and the task is migrated to the standby production unit.

[0050] After the repair is completed and the repairman clicks "Completion Confirmation" on the mobile device, the system immediately enters verification mode. The sensor array restarts high-frequency sampling to perform performance benchmark tests on the repaired components. If the vibration spectrum returns to below the normal baseline and the temperature gradient becomes gentler, the system determines that the fault has been officially resolved. At this point, the system automatically captures data such as the actual repair time, the batch number of spare parts used, and the skill level of the repairman, and feeds this data back to the maintenance knowledge base. The knowledge base uses reinforcement learning algorithms to dynamically adjust the recommended repair window duration and the accuracy of repair suggestions based on the actual repair results, achieving self-learning and evolution of the knowledge base.

[0051] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent manufacturing-oriented equipment health management and failure prediction and early warning, characterized in that, Includes the following steps: S1 utilizes a sensor array deployed in key parts of the manufacturing equipment to collect and process real-time operating status parameters of the equipment, obtaining multi-dimensional raw data; S2, acquire the time domain features, frequency domain features and spatial domain features of the multidimensional raw data, construct the equipment health status feature vector, and input the equipment health status feature vector into the preset fault prediction model; S3 identifies potential faults such as bearing wear or spindle imbalance in manufacturing equipment and outputs the fault type and its corresponding remaining safe operating time. S4. Based on the fault type and the remaining safe operating time of the equipment, a structured maintenance suggestion package is automatically calculated and generated by linking the maintenance knowledge base and the spare parts management database. S5, the structured maintenance suggestion package is pushed to the maintenance management module of the manufacturing execution system in real time. The manufacturing execution system parses the data items in the structured maintenance suggestion package and triggers the automated maintenance work order creation process and material reservation instructions. S6, start the production plan re-optimization algorithm, and under the premise of meeting the delivery date constraints and process constraints, migrate the subsequent production tasks of the affected equipment to the standby production unit, and reserve a certain maintenance execution window for the affected equipment to ensure that maintenance personnel and maintenance materials have arrived at the designated equipment location when the maintenance execution window is opened. 2.The smart manufacturing oriented device health management and failure prediction and early warning method according to claim 1, characterized in that, In step S1, the sensor array includes multiple triaxial vibration acceleration sensors, multiple high-precision infrared temperature sensors, and at least one Hall current sensor; wherein, the frequency response of the triaxial vibration acceleration sensor is within a preset frequency range, and the sampling frequency is set to a preset sampling frequency; the measurement accuracy of the high-precision infrared temperature sensor is better than a preset accuracy threshold; the Hall current sensor is connected in series at the output terminal of the spindle frequency converter to monitor the current fluctuation caused by changes in cutting load in real time. 3.The smart manufacturing-oriented device health management and failure prediction and early warning method according to claim 2, characterized in that, In step S3, the remaining safe operating time of the equipment is calculated based on the Weibull distribution function and combined with the real-time extracted feature vector offset to calculate the current reliability index of the equipment. When the reliability index is lower than the preset reliability threshold, it is determined to be in a warning state, and the specific remaining life value is output. 4.The smart manufacturing-oriented device health management and failure prediction and early warning method according to claim 3, characterized in that, In step S4, the structured maintenance recommendation package includes the severity level of the fault, the skill tags required for maintenance, the material codes of necessary spare parts, and the recommended maintenance window duration. 5.The smart manufacturing-oriented device health management and failure prediction and early warning method according to claim 4, characterized in that, The method for determining the recommended maintenance window duration is as follows: extract the historical average maintenance time for similar faults from the maintenance knowledge base, combine it with the current physical space layout of the equipment and the estimated time for spare parts to arrive, calculate the total time required to complete the maintenance, and add a preset proportion of redundant buffer time. The preset level in the severity level of the fault represents the highest priority and requires downtime processing within a preset time period.

6. The equipment health management and fault prediction and early warning method for intelligent manufacturing according to claim 5, characterized in that, In step S6, the production planning re-optimization algorithm adopts an improved genetic algorithm or ant colony algorithm. Its objective function is set to minimize the total order delay time and minimize the equipment switching cost. The constraints include the load rate of standby equipment not exceeding a preset load rate threshold, real-time availability of material supply, and consistency of product process path.

7. The equipment health management and fault prediction and early warning method for intelligent manufacturing according to claim 6, characterized in that, In step S6, the method for determining the migration of the production task is as follows: identify the process currently being executed by the affected equipment and determine the completion progress of the process; if the completion progress is less than a first preset progress threshold, the current process is immediately interrupted and migration is performed; if the completion progress is greater than a second preset progress threshold, the migration is performed after the current process is completed, and the material flow diagram in the manufacturing execution system is automatically updated.

8. The equipment health management and fault prediction and early warning method for intelligent manufacturing according to claim 7, characterized in that, Before launching the production planning re-optimization algorithm, a simulation verification step based on digital twins is also included: multiple candidate schemes generated by the algorithm are virtually run in a digital twin model to evaluate the impact of different scheduling strategies on output within a preset time period, and the optimal scheme is selected to perform task migration.

9. The equipment health management and fault prediction and early warning method for intelligent manufacturing according to claim 8, characterized in that, It also includes establishing a closed-loop feedback mechanism for equipment health status: after maintenance is completed, the equipment operation data is re-collected through the sensor array to verify whether the fault has been eliminated; the maintenance process data, actual time consumption data and spare parts consumption data are fed back to the maintenance knowledge base, and the maintenance knowledge base is self-learned and dynamically updated using reinforcement learning algorithms.

Images