Industrial equipment health state assessment and predictive maintenance method based on AI

Abstract

The invention discloses an AI-based industrial equipment health state assessment and predictive maintenance method, relates to the technical field of data processing, and can solve the problem of how to improve the reliability of health state assessment of industrial equipment in a working condition changing environment. Comprising the steps that multi-source monitoring data generated in the operation process of industrial equipment and corresponding working condition identification information are acquired; determining a health degradation trend index of the industrial equipment in the operation process according to the multi-source monitoring data; identifying a working condition switching event of the industrial equipment in the operation process according to the working condition identification information, and determining a working condition decoupling index of the industrial equipment in combination with the health degradation trend index; determining a credibility weight according to the working condition decoupling index, the multi-source monitoring data and a preset safety threshold; and determining a final health state evaluation result according to the credibility weight and the initial health state evaluation result, and generating a predictive maintenance strategy according to the final health state evaluation result.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to an AI-based method for assessing the health status of industrial equipment and predictive maintenance. Background Technology

[0002] With the deepening of intelligent transformation in the industrial manufacturing sector, ensuring the safe, stable, and efficient operation of critical equipment has become a core requirement, leading to widespread attention being paid to equipment health status assessment and predictive maintenance technologies. Currently, artificial intelligence (AI)-based methods have been applied in this field, typically using historical operating data to train models for automated assessment of equipment status and prediction of failure trends. However, industrial environments are complex, and equipment operating conditions frequently change. Existing assessment methods often struggle to fully guarantee the reliability of assessment results and the accuracy of maintenance decisions when faced with dynamically changing operating conditions. Summary of the Invention

[0003] To address the technical challenge of improving the reliability of industrial equipment health status assessment under changing operating conditions, this invention aims to provide an AI-based method for industrial equipment health status assessment and predictive maintenance. The specific technical solution adopted is as follows: In a first aspect, the present invention provides an AI-based method for assessing the health status of industrial equipment and predictive maintenance, comprising: acquiring multi-source monitoring data and corresponding operating condition identification information generated during the operation of industrial equipment; determining a health degradation trend index of the industrial equipment during operation based on the multi-source monitoring data; wherein the health degradation trend index is used to characterize the degree to which the operating status of the equipment deviates from the normal baseline; identifying operating condition switching events of the industrial equipment during operation based on the operating condition identification information, and determining an operating condition decoupling index of the industrial equipment based on the health degradation trend index and the operating condition switching events; wherein the operating condition decoupling index is used to characterize the degree of attenuation of the correlation between changes in equipment status and changes in operating conditions; determining a credibility weight based on the operating condition decoupling index, multi-source monitoring data, and a preset safety threshold; wherein the credibility weight is used to weight and modulate the initial health status assessment result output by a preset intelligent assessment model; determining a final health status assessment result based on the credibility weight and the initial health status assessment result, and generating a predictive maintenance strategy based on the final health status assessment result.

[0004] Secondly, the present invention provides an AI-based industrial equipment health status assessment and predictive maintenance system, including: a data acquisition module, a deviation trend assessment module, a decoupling degree assessment module, a credibility weight calculation module, and a status assessment and strategy maintenance module. The system comprises the following modules: a data acquisition module for acquiring multi-source monitoring data and corresponding operating condition identification information generated by industrial equipment during operation; a deviation trend assessment module for determining the health degradation trend index of industrial equipment during operation based on multi-source monitoring data, whereby the health degradation trend index characterizes the degree to which the equipment's operating status deviates from the normal baseline; a decoupling degree assessment module for identifying operating condition switching events of industrial equipment during operation based on operating condition identification information, and determining the operating condition decoupling index of industrial equipment based on the health degradation trend index and operating condition switching events, whereby the operating condition decoupling index characterizes the degree of attenuation of the correlation between equipment status changes and operating condition changes; a credibility weight calculation module for determining credibility weights based on the operating condition decoupling index, multi-source monitoring data, and preset safety thresholds, whereby the credibility weights are used to weight and modulate the initial health status assessment results output by the preset intelligent assessment model; and a status assessment and strategy maintenance module for determining the final health status assessment result based on the credibility weights and the initial health status assessment result, and generating predictive maintenance strategies based on the final health status assessment result.

[0005] Thirdly, the present invention provides an electronic device, comprising: a processor and a memory; wherein the memory is used to store one or more programs, the one or more programs including computer-executable instructions, and when the electronic device is running, the processor executes the computer-executable instructions stored in the memory to cause the electronic device to perform the AI-based industrial equipment health status assessment and predictive maintenance method as described in the first aspect and any possible implementation thereof.

[0006] This invention offers the following advantages: By introducing quantitative analysis of the health degradation trend index and the operating condition decoupling index, it dynamically identifies the true decline trend of equipment performance and decouples it from operating condition disturbances; furthermore, by combining the operational risk level, it generates credibility weights for modulating the initial evaluation results of the AI ​​model, effectively overcoming the problem of evaluation distortion caused by data distribution drift in static models; finally, it generates predictive maintenance strategies based on the weighted and modulated reliable health status, and continuously improves the model's adaptability through a closed-loop optimization mechanism. This method significantly improves the accuracy and robustness of industrial equipment health status assessment, as well as the scientific rigor and timeliness of predictive maintenance decisions, effectively avoiding missed or false alarms. Attached Figure Description

[0007] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the architecture of an AI-based industrial equipment health status assessment and predictive maintenance system provided in one embodiment of the present invention; Figure 2 This is a flowchart illustrating an AI-based method for assessing the health status of industrial equipment and predictive maintenance, provided as an embodiment of the present invention. Detailed Implementation

[0009] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0010] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0011] In all division and logarithmic operations involved in this invention, a smoothing mechanism is employed to prevent computer program crashes or invalid values ​​from being generated due to a zero denominator or zero input. Specifically, a correction factor ε, which is a very small positive number, is superimposed on the denominator term of the division operation or the argument term of the logarithmic function, for example, a value of 10 to the power of negative 5, thereby ensuring the robustness and feasibility of the algorithm under extreme conditions.

[0012] Unless otherwise specified, the normalization function Norm() mentioned in this invention uses maximum and minimum value normalization. The maximum and minimum values ​​are preset empirical extreme values ​​derived from a large amount of historical experimental data. If the calculation result exceeds the [0, 1] interval, it is restricted to the [0, 1] range by a truncation function (i.e., if the result is less than 0, it is taken as 0; if it is greater than 1, it is taken as 1) to eliminate the influence of outliers on the evaluation index.

[0013] The following description, in conjunction with the accompanying drawings, details a specific scheme for an AI-based industrial equipment health status assessment and predictive maintenance method provided by the present invention.

[0014] For example, such as Figure 1 The diagram shown is an architectural schematic of an AI-based industrial equipment health status assessment and predictive maintenance system (hereinafter referred to as the assessment and maintenance system) provided in an embodiment of the present invention. The assessment and maintenance system 10 includes: a data acquisition module 11, a deviation trend assessment module 12, a decoupling degree assessment module 13, a credibility weight calculation module 14, and a status assessment and strategy maintenance module 15. The modules are described in detail below: (1) Data acquisition module 11.

[0015] The data acquisition module 11 is responsible for collecting raw status and operating condition information from the industrial equipment operation site, providing basic data for all subsequent analysis and calculations.

[0016] Optionally, the data acquisition module 11 is used to acquire multi-source monitoring data and corresponding operating condition identification information generated by industrial equipment during operation.

[0017] Specifically, the data acquisition module 11 collects multi-source monitoring data reflecting the operating status of the equipment in real time through sensors (such as vibration sensors, temperature sensors, and current transformers) deployed in key parts of the industrial equipment. This data includes at least one of vibration signals, temperature parameters, current, voltage, and rotational speed. Simultaneously, the data acquisition module 11 obtains operating condition identification information corresponding to the timestamp of the monitoring data through the data interface of the equipment's host computer or programmable logic controller (PLC) or other control system. This information is used to characterize the load status, operating mode, or core process parameters of the equipment during operation.

[0018] Optionally, the data acquisition module 11 is also used to preprocess the multi-source monitoring data; wherein, the preprocessing includes at least one of data alignment, outlier removal and timestamp sorting.

[0019] The data acquisition module 11 aligns and packages the collected multi-source monitoring data and operating condition identification information in time to form a structured data stream, which is first output to the deviation trend assessment module 12 for trend analysis. At the same time, the operating condition identification information is synchronously provided to the decoupling degree assessment module 13.

[0020] (2) Deviation trend assessment module 12.

[0021] The deviation trend assessment module 12 is responsible for receiving multi-source monitoring data from the data acquisition module 11. By analyzing its continuous evolution characteristics in the time dimension, it quantifies the degradation trend of the equipment health status and provides key indicators for distinguishing between instantaneous disturbances and actual performance degradation.

[0022] Optionally, the deviation trend assessment module 12 is used to determine the health degradation trend index of industrial equipment based on the change characteristics of multi-source monitoring data over multiple consecutive time windows.

[0023] For example, the deviation trend assessment module 12 can be divided into a window segmentation submodule, a distribution difference calculation submodule, and a trend index determination submodule to respectively complete data segmentation, local feature extraction, and global trend synthesis, which are described below: (2.1) Window is divided into sub-modules.

[0024] Optionally, a window division submodule is used to divide multi-source monitoring data into multiple consecutive time windows according to a preset duration.

[0025] Specifically, the window segmentation submodule first buffers the input data stream. Then, based on a preset time length (e.g., 5 minutes) as the window size, it segments the continuous monitoring data sequence into a series of temporally continuous and partially overlapping or contiguous time window data blocks. Finally, these time window data blocks are output sequentially to the distribution difference calculation submodule.

[0026] (2.2) Distribution difference calculation submodule.

[0027] Optionally, the distribution difference calculation submodule is used to determine the window distribution difference amount within each time window based on the amplitude change characteristics and periodic change characteristics of each type of monitoring data sequence.

[0028] Specifically, the distribution difference calculation submodule first normalizes each type of monitoring data (such as vibration data) to eliminate the influence of dimensions. Then, it identifies the extreme points (maximum and minimum values) of the data sequence within the window and calculates the time interval between adjacent extreme points. Finally, based on the statistical characteristics of these time intervals (such as the reciprocal of the average interval) and the statistical characteristics of the amplitude ratio of the data sequence at adjacent sampling times (such as the average value), a window distribution difference quantity is calculated to characterize the rhythm and magnitude of amplitude abrupt changes within the window.

[0029] (2.3) Trend Index Determination Submodule.

[0030] Optionally, the trend index determination submodule is used to determine the health degradation trend index based on the degree of deviation and trend of the window distribution difference in multiple consecutive time windows relative to the historical normal reference value.

[0031] Specifically, the trend index determination submodule first retrieves the reference window distribution difference calculated during normal operation from the system's pre-stored historical data. Then, for each time window's window distribution difference output by the distribution difference calculation submodule, it calculates the degree of deviation from the reference value. Finally, it analyzes the changing trend of this deviation across multiple consecutive time windows (e.g., by combining the ratio of differences between adjacent windows with the deviation of the current window) to synthesize a single health degradation trend index. The larger this index, the more continuously and significantly the equipment's operating state deviates from the normal baseline, and the higher the probability of health degradation.

[0032] The health degradation trend index calculated by the deviation trend assessment module 12 is output to the decoupling degree assessment module 13 as one of the core inputs for assessing the impact of operating conditions.

[0033] (3) Decoupling degree assessment module 13.

[0034] The decoupling degree assessment module 13 is responsible for comprehensively analyzing the dynamic response relationship between operating condition switching events and health degradation trend index, aiming to isolate the state change part dominated by the performance degradation of the equipment itself.

[0035] Optionally, the decoupling degree assessment module 13 is used to identify operating condition switching events based on operating condition identification information and, in conjunction with the health degradation trend index, determine the operating condition decoupling index of industrial equipment.

[0036] Specifically, the decoupling degree assessment module 13 first identifies the moments when control commands or key process parameters undergo significant changes, i.e., operating condition switching events, from the operating condition identification information stream provided by the data acquisition module 11. Then, for each identified operating condition switching event, the module extracts the rate of change of operating condition parameters before and after the event and queries the rate of change of the health degradation trend index provided by the deviation trend assessment module 12 before and after the same moment. Furthermore, by calculating the degree of matching between these two rates of change (e.g., calculating their absolute difference), the state response offset under a single event is determined. Finally, the state response offsets corresponding to all operating condition switching events within the current analysis period are statistically analyzed, and combined with the frequency change trend of these events, the operating condition decoupling index is calculated. The higher this index, the less the equipment state evolution can be explained by operating condition changes, i.e., the more likely performance degradation is to dominate.

[0037] The decoupling index output by the decoupling degree assessment module 13 is transmitted to the credibility weight calculation module 14 to modulate the credibility of the assessment result.

[0038] (4) Credibility weight calculation module 14.

[0039] The credibility weight calculation module 14 is responsible for comprehensively evaluating the current operating risks and the autonomy of the state evolution of the equipment, and provides a dynamic credibility coefficient for the preliminary conclusions output by the preset intelligent evaluation model.

[0040] Optionally, the credibility weight calculation module 14 is used to determine the credibility weight based on the working condition decoupling index and the degree of proximity of multi-source monitoring data to a preset safety threshold.

[0041] Specifically, the credibility weight calculation module 14 first calculates the average amplitude of each type of data (such as current) within the current analysis period from the real-time monitoring data acquired by the data acquisition module 11. Then, it compares these average amplitudes with preset safety thresholds for each type of data, calculates the absolute value of the difference, and obtains the average level of the absolute values ​​of the differences across all categories to arrive at a comprehensive risk level value. The higher this value, the closer the average operating parameters of the equipment are to the safety red line. Finally, it combines the operating condition decoupling index provided by the decoupling degree assessment module 13 with a preset mapping relationship (such as an exponential decay relationship) to calculate the credibility weight. When both the operating condition decoupling index and the comprehensive risk level value are high, the calculated credibility weight is low, meaning that the reliability of the direct evaluation results of the AI ​​model decreases and should be treated with caution.

[0042] The credibility weights determined by the credibility weight calculation module 14 are output to the state assessment and strategy maintenance module 15 to correct the initial assessment results.

[0043] (5) Status assessment and strategy maintenance module 15.

[0044] The status assessment and strategy maintenance module 15 is the core of the assessment and maintenance system 10 for decision-making and self-optimization. It is responsible for comprehensively modulating the initial assessment results to generate the final health status conclusion and maintenance guidance, and completing the closed-loop update of the assessment model.

[0045] Optionally, the status assessment and strategy maintenance module 15 is used to determine the final health status assessment result based on the credibility weight and the initial health status assessment result output by the preset intelligent assessment model, and generate a predictive maintenance strategy accordingly; and to feed back the actual equipment operation data after the maintenance operation is performed according to the predictive maintenance strategy to the preset intelligent assessment model, and update and optimize the model.

[0046] For example, the state assessment and strategy maintenance module 15 can be divided into an assessment result modulation submodule, a strategy generation submodule, and a model optimization submodule to respectively complete result correction, strategy decision-making, and model iteration, which are described below: (5.1) Evaluation result modulation submodule.

[0047] Optionally, the evaluation result modulation submodule is used to determine the final health status evaluation result based on the initial health status evaluation result and the confidence weight.

[0048] Specifically, the evaluation result modulation submodule first invokes a preset intelligent evaluation model to calculate an initial health status assessment result based on the current multi-source monitoring data. This result is typically a quantitative score. Next, the evaluation result modulation submodule receives a confidence weight from the confidence weight calculation module 14. Finally, the evaluation result modulation submodule multiplies the initial health status assessment result by this confidence weight to obtain a weighted health status assessment value. This modulation process dynamically reduces the confidence level of the AI ​​model's direct output results when the operating conditions are highly decoupled and the operational risks are high, thus obtaining a more prudent and reliable final health status assessment result. The evaluation result modulation submodule outputs the modulated result to the strategy generation submodule.

[0049] (5.2) Strategy generation submodule.

[0050] Optionally, a strategy generation submodule is used to generate predictive maintenance strategies based on the final health status assessment results.

[0051] Specifically, the strategy generation submodule first receives a weighted health status assessment value from the evaluation result modulation submodule. To form a more comprehensive judgment, the strategy generation submodule can fuse this weighted value with at least one other source of health indicators (such as rule-based simple alarm status, periodic inspection records), for example, by using a weighted average algorithm to obtain a comprehensive health index.

[0052] Subsequently, the strategy generation submodule has pre-set maintenance strategy templates bound to different health level intervals (such as "Normal," "Observation," "Warning," and "Danger"). The strategy generation submodule matches the calculated comprehensive health index with these intervals, automatically generating corresponding predictive maintenance strategies. This strategy includes at least a suggested maintenance trigger time window and job priority, and can be expanded to include specific maintenance checklists or operation guidelines. The strategy generation submodule outputs this maintenance strategy to guide on-site operations, and simultaneously packages and sends relevant data from this evaluation cycle (including the final evaluation results, the data used, and the generated strategy) to the model optimization submodule.

[0053] (5.3) Model optimization submodule.

[0054] Optionally, the model optimization submodule is used to update and optimize the preset intelligent evaluation model using maintenance feedback data.

[0055] Specifically, the model optimization submodule receives data packets from the strategy generation submodule and continuously collects subsequent actual equipment operation data. After performing on-site maintenance operations according to the generated maintenance strategy, the model optimization submodule obtains the equipment monitoring data sequence before and after maintenance, and combines it with maintenance records (such as replacing a component) to accurately label the data for this period with health status data. This data with new labels constitutes a high-quality feedback dataset.

[0056] Subsequently, the model optimization submodule utilizes this feedback dataset to fine-tune or update the parameters of the preset intelligent evaluation model through incremental learning or periodic batch retraining algorithms. This process enables the model to learn the latest data characteristics of the equipment in the later stages of performance degradation and after maintenance intervention, thereby adapting to the data distribution drift of the equipment status and continuously improving the accuracy of the evaluation. The updated model will be used in subsequent evaluation cycles, forming a complete "evaluation-decision-feedback-optimization" closed loop.

[0057] The above provides an introduction to the assessment and maintenance system 10 and its included modules.

[0058] For example, such as Figure 2 The diagram shown is a flowchart illustrating an AI-based method for assessing the health status of industrial equipment and predictive maintenance, according to an embodiment of the present invention. The method includes the following steps: S201. Acquire multi-source monitoring data and corresponding operating condition identification information generated by industrial equipment during operation.

[0059] For example, this step can be performed by the data acquisition module 11 in the assessment and maintenance system 10 described above, and specifically includes the following steps: (1) Obtain multi-source monitoring data.

[0060] Specifically, the data acquisition module 11 collects multi-source monitoring data reflecting the operating status of the equipment in real time through sensors (such as vibration sensors, temperature sensors, current transformers, voltage transmitters, and speed encoders) deployed in key parts or related systems of industrial equipment. The multi-source monitoring data includes, but is not limited to, at least one of vibration signals, temperature parameters, current, voltage, and speed.

[0061] (2) Obtain the corresponding working condition identification information.

[0062] Specifically, the data acquisition module 11 synchronously acquires the operating condition identification information corresponding to the multi-source monitoring data in time through the data interface of the industrial equipment's control system (such as a programmable logic controller (PLC), a distributed control system (DCS), or a host computer). The operating condition identification information is used to characterize the current load status, operating mode, or core process parameter status of the equipment.

[0063] (3) Preprocess the multi-source monitoring data. Preprocessing includes at least one of the following: data alignment, outlier removal, and timestamp processing.

[0064] Optionally, the data acquisition module 11 cleans and organizes the collected raw multi-source monitoring data. Specifically, this includes: aligning the monitoring data from different sensors based on a unified time reference; using statistical methods (such as range filtering based on standard deviation) or rule-based logical judgment to remove obvious outlier sampling points; arranging all data in chronological order and organizing them into a structured data sequence with precise timestamps.

[0065] Thus, the data acquisition module 11, through the sensor network and control system interface, completes the synchronous acquisition, basic cleaning, and structured organization of multi-source heterogeneous data on the operating status of industrial equipment, providing a high-quality, spatiotemporally aligned input data stream for subsequent analysis.

[0066] S202. Based on multi-source monitoring data, determine the health degradation trend index of industrial equipment during operation. The health degradation trend index characterizes the degree to which the equipment's operating condition deviates from the normal baseline.

[0067] For example, this step can be performed by the deviation trend assessment module 12 in the assessment and maintenance system 10 described above. Specifically, the deviation trend assessment module 12 divides the continuous data sequence into multiple consecutive time windows according to a preset duration (e.g., 5 minutes). Next, within each time window, it analyzes the amplitude variation characteristics of each type of monitoring data sequence and the periodic variation characteristics reflected by the distribution of extreme points, calculating a window distribution difference quantity that comprehensively characterizes the data operation rhythm and the significance of amplitude mutations within that window. Then, based on the pre-stored reference window distribution difference quantity corresponding to the historical stage of normal equipment operation, it calculates the deviation degree of each time window difference quantity relative to the reference benchmark, analyzes the changing trend of this deviation degree within multiple consecutive time windows, and finally synthesizes a single health degradation trend index. It should be noted that the specific process of the aforementioned sub-steps can be found in S301-S304 below, and will not be repeated here.

[0068] In another possible implementation, when determining the health degradation trend index, the deviation trend assessment module 12 may not rely on fixed historical reference values. Instead, it may use the average difference in window distribution over a relatively long period of stable operation as a dynamic benchmark, and only calculate the deviation and trend of the current short-term window sequence relative to this dynamic benchmark to synthesize the index. This approach is suitable for scenarios where the operating condition baseline drifts slowly or where long-term stable historical data is lacking.

[0069] Therefore, the deviation trend assessment module 12 quantifies the continuous evolution of the equipment's operating status into a health degradation trend index that characterizes the persistence and significance of its deviation from the normal benchmark by dividing the multi-source monitoring data into time windows, extracting local features and synthesizing global trends, thus providing a key indicator for distinguishing between instantaneous disturbances and actual performance degradation.

[0070] S203. Identify operating condition switching events during the operation of industrial equipment based on operating condition identification information, and determine the operating condition decoupling index of the industrial equipment based on the health degradation trend index and the operating condition switching events. The operating condition decoupling index characterizes the degree of attenuation of the correlation between changes in equipment state and changes in operating conditions.

[0071] For example, this step can be performed by the decoupling degree assessment module 13 in the assessment and maintenance system 10 described above. Specifically, the decoupling degree assessment module 13 first identifies the moment when the control command or key process parameter changes significantly from the operating condition identification information flow, i.e., the operating condition switching event. For each identified operating condition switching event, the rate of change of the operating condition parameter before and after the event is extracted, and the rate of change of the health degradation trend index before and after the same moment is queried. By calculating the degree of matching between these two rates of change (such as the absolute value of the difference), the state response offset under a single event is determined. Finally, the state response offsets corresponding to all operating condition switching events within the current analysis period are statistically analyzed, and combined with the frequency change trend of these events, the operating condition decoupling index is calculated. It should be noted that the specific process of the aforementioned sub-steps can be found in S401-S404 below, and will not be repeated here.

[0072] In another possible implementation, when determining the decoupling index of industrial equipment, the decoupling degree assessment module 13 can not only compare the absolute differences in the rate of change when calculating a single offset, but also introduce the Pearson correlation coefficient or mutual information to measure the overall degree of correlation decay between the health degradation trend index sequence and the operating parameter sequence within the event window, and use this as part of the offset. The final index is obtained by combining the offsets and correlation decay degrees of multiple events.

[0073] Therefore, the decoupling degree assessment module 13 analyzes the degree of abnormality of the equipment state response under the working condition switching event, as well as the decoupling relationship between the state evolution trend and the driving force of the working condition change in time, and quantifies the degree of dominance of the equipment's own performance degradation in the state change, and generates a key working condition decoupling index.

[0074] S204. Determine the credibility weight based on the operating condition decoupling index, multi-source monitoring data, and preset safety thresholds. The credibility weight is used to weight and modulate the initial health status assessment results output by the preset intelligent assessment model.

[0075] For example, this step can be performed by the credibility weight calculation module 14 in the assessment and maintenance system 10 described above. Specifically, the credibility weight calculation module 14 first calculates the average amplitude of each type of monitoring data within the current analysis period from the real-time monitoring data. Then, it compares these average amplitudes with preset data safety thresholds for each type, calculates the absolute value of the difference, and obtains the average level of the absolute values ​​of the differences for all categories, thereby obtaining a comprehensive risk level value that characterizes the overall degree to which the equipment operating parameters are close to the safety red line. Finally, it calculates the credibility weight by combining the operating condition decoupling index with a preset mapping relationship (such as an exponential decay relationship). When both the operating condition decoupling index and the comprehensive risk level value are high, the calculated credibility weight is low, which means that the reliability of the direct assessment result of the AI ​​model decreases and needs to be treated with caution. It should be noted that the specific process of the aforementioned sub-steps can be found in S501-S503 below, and will not be repeated here.

[0076] In another possible implementation, when determining the credibility weight, the credibility weight calculation module 14 can also map the calculated comprehensive risk level value and the operating condition decoupling index to several preset discrete risk levels (such as "low", "medium", and "high"). Then, based on a predefined "risk level - credibility discount" lookup table, the final credibility weight is determined directly by looking up the table or through simple interpolation.

[0077] Therefore, the credibility weight calculation module 14 dynamically generates a weight coefficient to modulate the credibility of the initial evaluation results of the AI ​​model by comprehensively evaluating the risk level of the current operating parameters of the equipment and the degree to which the state evolution is dominated by its own degradation (i.e., the degree of decoupling of the operating conditions), providing an adjustment basis for generating more reliable final conclusions in the future.

[0078] S205. Based on the credibility weight and the initial health status assessment results, determine the final health status assessment results, and generate predictive maintenance strategies based on the final health status assessment results.

[0079] For example, this step can be performed by the status assessment and strategy maintenance module 15 in the assessment and maintenance system 10 described above, and specifically includes the following steps: (1) Determine the weighted health status assessment value based on the initial health status assessment results and the credibility weight.

[0080] Optionally, this step is performed by the evaluation result modulation submodule in the state assessment and strategy maintenance module 15.

[0081] Specifically, the evaluation result modulation submodule first calls a preset intelligent evaluation model (such as a fault prediction model based on deep learning) to calculate an initial health status evaluation result based on the current multi-source monitoring data. This result is typically a quantitative score (e.g., 0-100 points, with higher values ​​indicating healthier equipment). Then, the evaluation result modulation submodule receives the credibility weights from the credibility weight calculation module 14. w Finally, by using the initial health status assessment results... U Multiplying this by the confidence weight yields a weighted health status assessment value. This process dynamically reduces the confidence of the AI ​​model's direct output when the operating conditions are highly decoupled and the operational risks are high.

[0082] It should be noted that the preset intelligent assessment model is a pre-trained artificial intelligence model used for preliminary assessment of equipment health status. This model is preferably a deep learning-based regression or classification model, such as a Deep Neural Network (DNN), Convolutional Neural Network (CNN), or Long Short-Term Memory (LSTM). The model's input is a multi-source monitoring data sequence of a fixed time length (e.g., one analysis period), preprocessed and standardized by the data acquisition module 11. Its output is a quantified initial health status assessment result U (e.g., a continuous score between 0 and 100, or a discrete classification label representing states such as "healthy," "attention," or "abnormal"). The model utilizes a large amount of monitoring data accumulated during the equipment's historical operation and corresponding manually or automatically labeled health status labels (e.g., "normal" or "faulty" and their severity) for supervised training, learning the mapping relationship from multi-source time-series data to equipment health status. After deployment, this model serves as a fixed component in the assessment process, providing initial judgments, and its output will be dynamically modulated by subsequent confidence weights.

[0083] (2) Based on the weighted health status assessment value and at least one other health indicator, the comprehensive health index is obtained.

[0084] Furthermore, the evaluation result modulation submodule integrates the aforementioned weighted health status assessment value with other possible health indicators (such as simple alarm status based on fixed rules, and scores from regular manual inspection records). Algorithms such as weighted averaging and Bayesian fusion can be used to calculate a more comprehensive and robust overall health index.

[0085] (3) Determine the corresponding predictive maintenance strategy based on the preset health level range of the comprehensive health index; wherein the predictive maintenance strategy includes maintenance triggering time and maintenance priority.

[0086] Optionally, the strategy generation submodule in the status assessment and strategy maintenance module 15 has pre-set maintenance strategy templates bound to different health level ranges (such as "normal", "observation", "warning", and "danger"). The strategy generation submodule matches the calculated comprehensive health index with these ranges and automatically generates corresponding predictive maintenance strategies. This strategy includes at least a suggested maintenance trigger time window and job priority, and can be expanded to include specific maintenance checklists or operation guidelines.

[0087] (4) Feed the actual equipment operation data after performing maintenance operations according to the predictive maintenance strategy back to the preset intelligent evaluation model and update and optimize the preset intelligent evaluation model.

[0088] Optionally, the model optimization submodule in the condition assessment and strategy maintenance module 15 performs this step. After performing on-site maintenance operations according to the generated maintenance strategy, the model optimization submodule acquires the equipment monitoring data sequence before and after maintenance, and combines it with maintenance records (such as replacing a component) to accurately label the data for this period. This data with new labels constitutes a high-quality feedback dataset. The model optimization submodule uses this dataset to update the parameters of the preset intelligent assessment model through incremental learning or periodic batch retraining algorithms, enabling the model to adapt to the data distribution drift of the equipment condition and continuously improve the assessment accuracy.

[0089] Thus, the status assessment and strategy maintenance module 15 achieves full-process automation and intelligence from assessment and decision-making to self-improvement by weighting the initial AI assessment results with credibility, integrating multi-source indicators to generate a comprehensive health index, formulating differentiated maintenance strategies based on health levels, and using maintenance feedback data to optimize the model in a closed loop.

[0090] Based on the above technical solution, this invention introduces quantitative analysis of a health degradation trend index and an operating condition decoupling index to dynamically identify the true decline trend of equipment performance and decouple it from operating condition disturbances. Furthermore, by combining the operational risk level, it generates credibility weights for modulating the initial evaluation results of the AI ​​model, effectively overcoming the problem of evaluation distortion in static models under data distribution drift. Finally, it generates predictive maintenance strategies based on the weighted and modulated reliable health status and continuously improves the model's adaptability through a closed-loop optimization mechanism. This method significantly improves the accuracy and robustness of industrial equipment health status assessment, as well as the scientific rigor and timeliness of predictive maintenance decisions, effectively avoiding missed or false alarms.

[0091] For example, in another AI-based method for assessing the health status of industrial equipment and predictive maintenance provided in one embodiment of the present invention, the health degradation trend index of industrial equipment during operation is determined based on multi-source monitoring data, specifically including the following steps: S301. Divide the multi-source monitoring data into multiple continuous time windows according to a preset duration.

[0092] For example, this step can be performed by the window partitioning submodule in the deviation trend assessment module 12.

[0093] Specifically, the window segmentation submodule caches the multi-source monitoring data sequence from the data acquisition module 11, arranged in chronological order. Based on a preset fixed duration (e.g., 5 minutes), the continuous monitoring data stream is divided into a series of temporally continuous and contiguous time window data blocks. Each time window contains sampling points for all categories of monitoring data within that time period.

[0094] It should be noted that the time windows involved in the embodiments of the present invention are divided into two categories: 1. Current analysis time window: refers to the sequence of continuous time windows used for real-time health status assessment, which is divided from the current real-time monitoring data stream; 2. Historical reference time window: refers to the time window sequence used to establish health benchmarks, which is divided from the monitoring data of the historical normal operation phase.

[0095] For the current analysis, the window segmentation submodule segments the current monitoring data in real time; for historical reference, the system pre-divides time windows from historical normal operation data and calculates statistical characteristics.

[0096] S302. Within each time window, based on the amplitude and periodicity characteristics of each type of monitoring data sequence, determine the window distribution difference for each type of monitoring data within each time window. The window distribution difference is used to comprehensively characterize the significance of changes in the rhythm and amplitude of the monitoring data within a short time window.

[0097] For example, this step can be performed by the distribution difference calculation submodule in the deviation trend assessment module 12, specifically including the following steps: (1) Normalize the monitoring data sequence.

[0098] First, the distribution difference calculation submodule performs normalization preprocessing on each type of monitoring data input to the current time window. The raw amplitude data is mapped to the [0, 1] interval to eliminate the influence of different physical dimensions on subsequent calculations. After normalization, the amplitude of the monitoring data at each sampling time is denoted as... hIt should be noted that the current time window specifically refers to the data set corresponding to the latest segment of the continuous monitoring data stream, defined according to a preset duration (e.g., 5 minutes), during which a health status assessment and analysis is being conducted. This window contains the sampled values ​​of all monitoring indicators within this time period, arranged in chronological order, and is the basic data unit for calculating the window distribution difference.

[0099] (2) Identify the extreme points in the normalized monitoring data sequence and determine the time interval between adjacent extreme points.

[0100] Next, the distribution difference calculation submodule identifies all local extrema (including maxima and minima) in the normalized data sequence. It calculates the time interval between any two adjacent extrema, denoted as . The average time interval is obtained by averaging all such intervals within the current time window. .

[0101] (3) Determine the window distribution difference based on the statistical characteristics of the time interval between adjacent extreme points and the statistical characteristics of the amplitude ratio of the monitoring data sequence at adjacent sampling times.

[0102] Finally, the distribution difference calculation submodule integrates the time interval characteristics and amplitude change characteristics to calculate the window distribution difference of this type of monitoring data within the current time window. f For example, it can be calculated using the following formula: in, This represents the average time interval between adjacent extreme points of this type of monitoring data within the current time window. n represents the total number of sampling points for this type of monitoring data within the current time window. , These represent the amplitudes of the monitoring data of this type at the i-th and i+1-th sampling times within the current time window after normalization. This indicates that the average value is calculated for all categories of monitoring data (such as vibration, rotational speed, and power) within the current time window.

[0103] It is understandable that in the above formula The evaluation assesses the significance of changes in the rhythm (the shorter the interval, the faster the rhythm change). The evaluation assesses the drastic change in amplitude between adjacent time points. Multiplying the two values ​​and then averaging them across all monitoring data categories yields the window distribution variance, which characterizes the degree of abrupt changes in the overall state within that time window. f .

[0104] It should be noted that the window distribution difference calculated by the above formula... fIts value includes the reciprocal of the time interval ( It has the dimension of time [1 / T]. This difference in dimensions does not affect subsequent analysis because in step S303, the time windows of each time window... f The values ​​will be normalized together with historical reference values, and converted into a dimensionless relative scale. s This normalization process unifies all comparison benchmarks, making them consistent across different benchmarks. s The deviation calculations and trend analyses of the values ​​are performed within a dimensionless, comparable numerical system, thereby eliminating the influence of dimensions on the final health degradation trend index.

[0105] S303. Based on the reference window distribution difference corresponding to the normal operation history of the industrial equipment, determine the degree of deviation of the window distribution difference in each time window from the reference window distribution difference.

[0106] For example, this step can be performed by the trend index determination submodule in the deviation trend assessment module 12.

[0107] Specifically, the trend index determination submodule retrieves data from the system storage for multiple time windows during the device's historical normal operation phase (such as a stable, fault-free operation period). Following method S302, it calculates the window distribution difference for each time window, then takes the arithmetic mean of these window distribution differences as the reference window distribution difference, and normalizes it, denoted as... s For the current time window distribution difference calculated in step S302. f The trend index determination submodule first normalizes it, denoted as . s Then calculate the degree of deviation of the window, i.e., | s - s This deviation directly quantifies the difference between the current window's operational characteristics and its historical health benchmark.

[0108] S304. Determine the health degradation trend index based on the changing trend of the degree of deviation in multiple consecutive time windows.

[0109] For example, when the trend index determination submodule determines the health degradation trend index, it specifically includes the following steps: (1) Determine the first ratio of the window distribution difference of each time window to the window distribution difference of the adjacent previous window.

[0110] Specifically, the trend index determination submodule selects continuous data including the current window. m The first time window. For the first c windows ( c ≥2), calculate its normalized window distribution difference. With the previous window The ratio, which is the first ratio, reflects the amplified trend of the state difference between adjacent windows.

[0111] (2) Determine the health degradation trend index based on the first ratio and the degree of deviation of multiple consecutive time windows.

[0112] Furthermore, the trend index determination submodule synthesizes the health degradation trend index using the following formula. : Where m is the total number of selected continuous time windows. For the first c The normalized window distribution variance of each window. s ′ is the reference baseline value. α It is a very small non-zero constant (for example, the value is 0.001).

[0113] It should be noted that in the above formula The first was measured c The absolute degree to which the state of each window deviates from the baseline. The relative growth trend of this deviation in the next window is measured. The "deviation" of each window is multiplied by its subsequent "growth trend," and then the average is taken over all windows to obtain a comprehensive index. This health degradation trend index r The larger the value, the more it indicates that the equipment's operating status not only continuously deviates from the normal baseline, but also exhibits a stable or accelerating evolutionary behavior, significantly increasing the possibility of equipment health degradation.

[0114] Based on the above technical solution, the embodiments of the present invention divide the equipment operation data into time windows, calculate the window distribution difference by fusing amplitude and rhythm characteristics within each window, and synthesize a health degradation trend index that can sensitively capture the slow and continuous performance degradation process by analyzing the deviation of multiple consecutive window differences from historical benchmarks and their evolution trends. This provides a core basis for accurately distinguishing instantaneous disturbances from real health degradation.

[0115] For example, in another AI-based industrial equipment health status assessment and predictive maintenance method provided in one embodiment of the present invention, the method identifies operating condition switching events of industrial equipment during operation based on operating condition identification information, and determines the operating condition decoupling index of industrial equipment based on health degradation trend index and operating condition switching events. Specifically, the method includes the following steps: S401. Identify the operating condition switching events of industrial equipment during operation based on the operating condition identification information.

[0116] In this step, the decoupling degree assessment module 13 analyzes the change records of control commands or key process parameters in real time from the operating condition identification information data stream provided by the data acquisition module 11. By monitoring whether variables characterizing equipment load status, operating mode, or core process parameters (e.g., motor power setpoint, production line speed command, processing pressure threshold) undergo step changes or continuous significant changes exceeding preset sensitivity, the start and end times of each operating condition switch are identified.

[0117] For example, the aforementioned preset sensitivity can be set according to the process control accuracy. For speed control, an event with a change exceeding 5% of the rated value can be identified as a condition switch. The decoupling degree evaluation module 13 generates a record containing a switching timestamp and parameter values ​​before and after the change for each identified event, as input for subsequent analysis.

[0118] S402. For each operating condition switching event, determine the rate of change of operating condition parameters before and after the operating condition switching, as well as the rate of change of the health degradation trend index before and after the operating condition switching.

[0119] Furthermore, the decoupling degree evaluation module 13 targets the first identified by S401. j The next operating condition switch event is processed. First, the normalized operating condition parameter values ​​before and after the switch are obtained and denoted as follows: and Calculate the rate of change of operating parameters. Its value is the absolute value of the difference between the preceding and following parameters, i.e. Simultaneously, the health degradation trend index calculated before and after the same time (i.e., within the time period defined by this operating condition switch) is obtained from the deviation trend assessment module 12 and recorded as follows: and Calculate the rate of change of the health degradation trend index. An example calculation formula is as follows: in, and They represent the first j Health degradation trend index before and after the next operating condition switch; α This is a very small non-zero constant (example value is 0.001). It can be understood that the numerator in this formula... It quantifies the absolute change of the index before and after the change in operating conditions, plus α Ensure it is positive; denominator This absolute change is then converted into a relative rate of change relative to the index level after the switch. The larger the value, the more drastic the relative change in the condition index before and after this condition switch.

[0120] S403. Determine the state response offset under a single operating condition switching event based on the degree of matching between the rate of change of operating condition parameters and the rate of change of health degradation trend index.

[0121] For example, the decoupling degree evaluation module 13 calculates the first decoupling degree using the following formula. j State response offset under secondary operating condition switching event : in, , These represent the rate of change of the health degradation trend index and the rate of change of the operating parameters calculated in S402, respectively. and They represent the first j Health degradation trend index before and after the next operating condition switch; and They represent the first j Second and third j +1 operating condition parameter value after operating condition switch; α It is a very small non-zero constant (for example, the value is 0.001).

[0122] It should be noted that the first term in the above formula... This measure assesses the degree of matching between the rate of change of state and the rate of change of operating condition in terms of amplitude; a larger difference indicates a greater mismatch between the state response and the operating condition drive in terms of the intensity of change. The second term... This measures the consistency in trend between the percentage change in the state index and the percentage change in the operating parameters; a larger difference indicates a greater asynchrony in their change rates. Multiplying these two values ​​yields the state response offset. The larger the value, the less the changes in the equipment's operating status can be reasonably explained by the current switching of operating conditions and its subsequent changes; that is, the more obvious the deviation.

[0123] S404. Obtain the state response offset corresponding to multiple operating condition switching events within the current analysis period, and determine the operating condition decoupling index by combining the frequency change trend of multiple operating condition switching events.

[0124] Finally, the decoupling degree evaluation module 13 comprehensively processes the state response offsets within the current analysis period (e.g., a relatively long period containing multiple consecutive operating condition switches, such as 30 minutes). First, the analysis period is divided into multiple consecutive short-term time windows (e.g., each window is 5 minutes, consistent with the definition in S301). For each time window, the state response offsets corresponding to all operating condition switching events occurring within that window are calculated. The average value of the average state response offset for that time window is obtained, denoted as . At the same time, the number of operating condition changes that occurred within each time window is counted.

[0125] Then, the decoupling degree evaluation module 13 evaluates the average state response offset sequence of multiple consecutive time windows respectively. The sequence of operating condition switching times and the sequence of operating condition switching times are linearly fitted using the least squares method according to the time window order, resulting in two fitted curves. The slope of the change between adjacent time windows is extracted from the fitted curves to form the slope sequence of the average state response offset. The slope sequence of the number of operating condition switching }

[0126] Next, data with similar or identical operating conditions and ranges of change are filtered from historical data (i.e., Multiple operating condition switching events where the difference is less than a preset threshold (e.g., the difference in change after normalization is less than 0.05). These events are arranged in chronological order of occurrence, and their corresponding state response offsets (calculated using formula S403) are recorded as follows: Calculate the ratio of two consecutive offsets, i.e. Take the arithmetic mean of these ratios, denoted as . .

[0127] Furthermore, the operating condition decoupling index is calculated using the following formula. F : Where N represents the number of slope pairs involved in the fitting; and They represent the first z The slope of the trend of state response offset at each location and the slope of the trend of the number of operating condition switching; and It is the state response offset corresponding to two adjacent switching events with the same (or similar) change amplitude of operating conditions, selected from historical data; It represents the average of the ratio of the state response offsets corresponding to two adjacent switching events under the same operating condition change range; α It is a very small non-zero constant (for example, the value is 0.001).

[0128] It should be noted that in the above formula This term quantifies the synchronicity between the overall trend of state shifts and the trend of operating condition switching frequency over a long time scale. The larger the value, the less synchronized the state evolution trend is with the rhythm of operating condition changes, meaning that state changes are more independent of operating condition-driven processes. The term quantifies the cumulative growth trend of equipment state response deviation under identical operating conditions. A ratio greater than 1 and continuously increasing indicates that even with identical external conditions, equipment state anomalies are constantly accumulating and worsening, providing direct evidence of performance degradation. Multiplying the two results in a larger operating condition decoupling index F, indicating that the equipment state evolution is not only becoming increasingly detached from the influence of operating condition changes, but also that its degree of anomaly is continuously worsening under the same external stimuli. This strongly suggests that the equipment's own performance degradation has become the dominant factor in state changes.

[0129] Based on the above technical solution, this invention accurately identifies operating condition switching events, quantifies the matching degree between the equipment state response and the operating condition drive in terms of the magnitude and rhythm of changes during a single switch, and obtains the state response offset. Furthermore, it analyzes the synchronicity between the state offset trend and the operating condition change trend over a long time scale, and combines this with the cumulative growth trend of the offset under the same excitation to finally synthesize an operating condition decoupling index. This index effectively isolates the influence of operating condition disturbances and quantifies the dominance of equipment performance degradation in state evolution, thus providing crucial decoupling information for subsequent accurate assessment of health status.

[0130] For example, in another AI-based industrial equipment health status assessment and predictive maintenance method provided in one embodiment of the present invention, the confidence weight is determined based on the operating condition decoupling index, multi-source monitoring data, and preset safety thresholds, specifically including the following steps: S501. Determine the absolute value of the difference between the average amplitude of each type of monitoring data and the preset safety threshold within the current analysis period.

[0131] Specifically, the credibility weight calculation module 14 obtains multi-source monitoring data from the data acquisition module 11 within the current analysis time period (e.g., the most recent evaluation cycle, such as 30 minutes). For each type of monitoring data (e.g., vibration, current, temperature), it calculates the arithmetic mean of the amplitudes of all sampling points within that time period, denoted as... Simultaneously, the preset safety thresholds corresponding to each type of monitoring data are obtained from the system configuration or equipment technical specifications. H This threshold represents the upper (or lower, in this case, the upper) limit allowed for the safe operation of the equipment. For example, for motor winding temperature, there is a preset safety threshold. H The threshold can be set to the maximum permissible temperature of the insulation material (e.g., 155°C); for vibration intensity, it can be set to the upper limit of the "good" zone for the corresponding equipment type in ISO 10816 standard. Threshold setting rules are typically based on the equipment manufacturer's rated parameters, international / industry standards, or statistical safety margins from long-term historical operating data (e.g., taking the 99.5th percentile of the historical normal value distribution). Furthermore, the absolute value of the difference between the average amplitude of each type of monitoring data and the safety threshold for that type of data is calculated, i.e., | - H |

[0132] S502. Determine the comprehensive risk characterization value of multi-source monitoring data based on the average of the absolute values ​​of the differences among all categories of monitoring data.

[0133] Furthermore, the credibility weight calculation module 14 gathers all K The absolute values ​​of the differences obtained from the monitoring data are used to calculate the arithmetic mean of these absolute values ​​to obtain the comprehensive risk characterization value R, which represents the degree to which the overall operating parameters of the equipment approach the safety red line. An example calculation formula is as follows: in, Indicates the first k The average amplitude of the monitoring data within the current analysis period; Indicates the first k Preset safety thresholds corresponding to the monitoring data; α K is a very small non-zero constant (0.001 for example); K represents the total number of categories of monitoring data.

[0134] It should be noted that, based on the above formula, for each type of data, the absolute value of the difference... The smaller the value, the closer the average of this type of operating parameter is to the safety upper limit, and the higher the risk. A comprehensive risk characterization value is obtained by averaging the risk contributions of all categories. R . R The smaller the value, the closer the average operating parameters of the equipment are to their respective safety thresholds, meaning the higher the overall operational risk level.

[0135] S503. Determine the credibility weight based on the operating condition decoupling index and the comprehensive risk characterization value.

[0136] For example, the credibility weight calculation module 14 receives the operating condition decoupling index from the decoupling degree evaluation module 13. F (Calculated from S404), and combined with the comprehensive risk characterization value calculated from S502. R The credibility weight is calculated using the following formula. : in, This represents the natural exponential function. In this formula, the confidence weight... w It is the operating condition decoupling index F The reciprocal of the comprehensive risk characterization value 1 / R The negative exponential function of the product. When the operating condition is decoupled, the exponential function... F Larger (degradation is the dominant factor) and comprehensive risk characterization value RWhen the product is relatively small (operating parameters are close to the safety threshold), The value of will be very large, causing the value of the exponential function to decrease sharply, i.e., the confidence weight. w Approaching 0. Conversely, when the equipment state is mainly driven by operating conditions ( F Small) and operating parameters are far from the safety threshold ( R When the product is large, the confidence weight is small. w The weight approaches 1. Therefore, this weight dynamically reflects the reliability of the initial health status assessment results based on the AI ​​model under the combined influence of current operational risks and the degree of degradation dominance. The lower the weight, the lower the confidence level of directly using the AI ​​model's output results, requiring more caution or stronger modulation.

[0137] Based on the above technical solution, this embodiment of the invention dynamically generates an exponentially decaying confidence weight by calculating the degree to which the overall operating parameters of the equipment approach the safety threshold (comprehensive risk characterization value) and combining it with a condition decoupling index that reflects the degree of performance degradation dominance. This weight can intelligently identify key scenarios where "high operating risk" and "high degradation autonomy" coexist, and automatically reduce the confidence level of the initial evaluation results of the AI ​​model in such scenarios, thereby providing a scientific basis for modulation to generate more reliable and robust final health status evaluation results.

[0138] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0139] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An AI-based method for assessing the health status of industrial equipment and predictive maintenance, characterized in that, The method includes: Acquire multi-source monitoring data and corresponding operating condition identification information generated by industrial equipment during operation; Based on the multi-source monitoring data, a health degradation trend index is determined for the industrial equipment during operation; wherein, the health degradation trend index is used to characterize the degree to which the equipment's operating status deviates from the normal baseline; The operating condition switching events of the industrial equipment during operation are identified based on the operating condition identification information, and the operating condition decoupling index of the industrial equipment is determined based on the health degradation trend index and the operating condition switching events; wherein, the operating condition decoupling index is used to characterize the degree of attenuation of the correlation between equipment state changes and operating condition changes. The credibility weight is determined based on the operating condition decoupling index, the multi-source monitoring data, and the preset safety threshold; wherein, the credibility weight is used to weight and modulate the initial health status assessment results output by the preset intelligent assessment model. Based on the credibility weight and the initial health status assessment result, the final health status assessment result is determined, and a predictive maintenance strategy is generated based on the final health status assessment result.

2. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 1, characterized in that, Based on the multi-source monitoring data, the health degradation trend index of the industrial equipment during operation is determined, specifically including: The multi-source monitoring data is divided into multiple consecutive time windows according to a preset duration; Within each time window, based on the amplitude and periodicity variation characteristics of each type of monitoring data sequence, the window distribution difference of each type of monitoring data within each time window is determined; wherein, the window distribution difference is used to comprehensively characterize the degree of significance of the changes in the operating rhythm and amplitude of the monitoring data within a short time window; Based on the reference window distribution difference corresponding to the normal operation history of the industrial equipment, determine the degree of deviation of the window distribution difference within each time window from the reference window distribution difference. The health degradation trend index is determined based on the changing trends of the degree of deviation in multiple consecutive time windows.

3. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 2, characterized in that, Within each time window, based on the amplitude and periodicity characteristics of each type of monitoring data sequence, the window distribution difference of each type of monitoring data within each time window is determined, specifically including: The monitoring data sequence is normalized. Identify extreme points in the normalized monitoring data sequence and determine the time interval between adjacent extreme points; The window distribution difference is determined based on the statistical characteristics of the time interval between adjacent extreme points and the statistical characteristics of the amplitude ratio of the monitoring data sequence at adjacent sampling times.

4. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 2, characterized in that, The health degradation trend index is determined based on the changing trends of the degree of deviation over multiple consecutive time windows, specifically including: Determine a first ratio of the window distribution difference of each time window to the window distribution difference of the adjacent preceding window; The health degradation trend index is determined based on the first ratio of multiple consecutive time windows and the degree of deviation.

5. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 1, characterized in that, Based on the health degradation trend index and the operating condition switching event, the operating condition decoupling index of the industrial equipment is determined, specifically including: For each operating condition switching event, determine the rate of change of operating condition parameters before and after the operating condition switching, as well as the rate of change of the health degradation trend index before and after the operating condition switching; Based on the degree of matching between the rate of change of the operating condition parameters and the rate of change of the health degradation trend index, the state response offset under a single operating condition switching event is determined. Obtain the state response offset corresponding to multiple operating condition switching events within the current analysis period, and determine the operating condition decoupling index by combining the frequency change trend of the multiple operating condition switching events.

6. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 1, characterized in that, The reliability weight is determined based on the operating condition decoupling index, the multi-source monitoring data, and the preset safety threshold, specifically including: Determine the absolute value of the difference between the average amplitude of each type of monitoring data within the current analysis period and the preset safety threshold; The comprehensive risk characterization value of the multi-source monitoring data is determined based on the average of the absolute values ​​of the differences among all categories of monitoring data. The credibility weight is determined based on the operating condition decoupling index and the comprehensive risk characterization value.

7. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 1, characterized in that, Based on the confidence weight and the initial health status assessment result, a final health status assessment result is determined, and a predictive maintenance strategy is generated based on the final health status assessment result, specifically including: The weighted health status assessment value is determined based on the initial health status assessment results and the confidence weight. A comprehensive health index is obtained based on the weighted health status assessment value and at least one other health indicator; Based on the preset health level range of the comprehensive health index, a corresponding predictive maintenance strategy is determined; wherein, the predictive maintenance strategy includes maintenance triggering timing and maintenance priority.

8. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 1, characterized in that, Acquire multi-source monitoring data and corresponding operating condition identification information generated by industrial equipment during operation, specifically including: Sensors deployed on the industrial equipment are used to collect multi-source monitoring data reflecting the equipment's operating status in real time; wherein the multi-source monitoring data includes at least one of the following: vibration signal, temperature parameter, current, voltage, and rotational speed; The operating condition identification information corresponding to the time of the multi-source monitoring data is synchronously acquired from the control system of the industrial equipment; wherein, the operating condition identification information is used to characterize the current load status, operating mode or process parameters of the equipment.

9. The AI-based industrial equipment health status assessment and predictive maintenance method according to claim 8, characterized in that, After acquiring multi-source monitoring data and corresponding operating condition identification information generated by industrial equipment during operation, the method further includes: The multi-source monitoring data is preprocessed; wherein the preprocessing includes at least one of data alignment, outlier removal, and timestamp processing.

10. The AI-based industrial equipment health status assessment and predictive maintenance method according to any one of claims 1-9, characterized in that, The method further includes: The actual equipment operation data after performing maintenance operations according to the predictive maintenance strategy is fed back to the preset intelligent evaluation model to update and optimize the preset intelligent evaluation model.

Images