Air compressor regulation and control system and method based on data driving

Abstract

The invention discloses an air compressor regulation and control system and method based on data driving, and relates to the technical field of air compressor regulation and control. Physical parameters of an air compressor and equipment response data corresponding to each control instruction are synchronously obtained; based on the historical normal operation data, establishing equipment reference response characteristics corresponding to different control instructions; the method comprises the following steps: establishing a reference response characteristic, comparing real-time response data with the reference response characteristic, identifying response drift, delay elongation, feedback structure change and behavior deviation under similar instructions when the control logic is not changed, and outputting a dynamic evolution track of a deviation characteristic, according to the method, by establishing the reference response characteristic and comparing the reference response characteristic with the real-time response data, identifying the early response deviation of the equipment; potential faults are found in advance, and stable operation of equipment is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of air compressor control, and in particular relates to a data-driven air compressor control system and method. Background Art

[0002] An air compressor is a device that converts mechanical energy into gas pressure energy and is widely used in industry, manufacturing, construction, medical care and other fields.

[0003] During air compressor operation, traditional control methods struggle to accurately detect potential equipment failures in their early stages. Due to long-term operation, the equipment may experience structural fatigue, control link anomalies, or internal medium degradation. These problems often don't initially cause physical parameters to exceed their limits, but they can cause subtle changes in the equipment's response, such as response drift and extended delays. Failure to promptly identify these early signs and adjust control commands can lead to more serious failures, impacting the normal operation and service life of the air compressor. Therefore, accurately identifying early equipment anomalies and implementing control measures based on data-driven methods is key to ensuring stable and efficient air compressor operation. Summary of the Invention

[0004] The purpose of the present invention is to provide a data-driven air compressor control system and method, which solves the technical problem in the prior art that it is impossible to accurately identify early abnormalities of equipment and perform control based on data driving.

[0005] The data-driven air compressor control method includes: Synchronously obtain the physical parameters of the air compressor and the device response data corresponding to each control instruction; Based on historical normal operation data, establish the equipment benchmark response characteristics corresponding to different control instructions; Compare real-time response data with baseline response characteristics to identify response drift, delay extension, feedback structure changes, and behavioral deviations under similar instructions when the control logic remains unchanged, and output the dynamic evolution trajectory of the deviation characteristics; When the behavioral deviation exceeds the set range, the transmission stages of parameter micro-fluctuations are first divided based on the key turning points of the deviation evolution trajectory. The influence relationship between the parameters in each stage is then checked to see whether it reproduces the transmission pattern of the historical degradation process under similar working conditions. If the sequence of change and intensity distribution of the parameter influence relationship in each stage are consistent with the historical characteristics, even if the physical parameters are within the limit, it is still determined that the equipment has early signs of structural fatigue, control link abnormality, or internal medium degradation. The control instructions are adjusted according to the risk assessment results, and the new instructions and response data are fed back to the expected response model unit to update the baseline characteristics.

[0006] Furthermore, after determining that the device has early signs of failure, it also includes: Obtain the parameter influence relationship in each conduction stage that is inconsistent with the historical characteristic change order and mark it as an abnormal action point; Taking the abnormal action point as the starting point, obtain the response time difference sequence of the parameter after the control command is issued, and compare it with the characteristics of the time difference sequence of historical normal response; If the interval fluctuation amplitude of a certain link in the time difference sequence exceeds the normal range of similar working conditions, intercept the equipment operation soundprint or vibration spectrum fragment corresponding to this link and compare the resonance peak with the soundprint / spectrum of the same link at the early stage of historical degradation; If the resonance peak shift does not reach the warning threshold, adjust the analysis dimension of the parameter interaction relationship and recheck the transition order; After each round of sorting, count the number of abnormal action points to see if they have decreased. If they have not decreased for two consecutive times, use the typical deviation transmission path of the same model of equipment in the same degradation scenario as a reference and check the consistency of the current time difference sequence with the reference path step by step. Until the number of abnormal action points is lower than the set threshold, the final located deviation source link and verification parameters are associated with the early sign judgment results and updated to the baseline feature library of the expected response model unit.

[0007] Furthermore, in the early stage of parameter micro-fluctuation transmission, the following is also included: Obtain parameter combinations that do not match the intensity distribution of the historical degradation conduction law in each stage, and align the control command curves corresponding to these parameters with the equipment response curves; If the number of curve intersections exceeds the preset range for similar operating conditions in multiple consecutive fluctuation cycles after alignment, intercept the equipment operating vibration waveform in the area with dense intersections and compare it with the vibration waveform in the normal state at the same stage for trough spacing; According to the difference in trough spacing, adjust the sampling interval of the parameter combination and recheck the intensity distribution until it conforms to the historical characteristics. Archive the adjusted sampling interval together with the vibration waveform characteristics.

[0008] Furthermore, in the mid-term of verifying the impact of parameters, it also includes: Find the parameter combination with the largest deviation from the historical change order in each stage, and calculate the interval between the instruction triggering time and the response extreme value appearance time corresponding to this group; If the dispersion of the interval value exceeds the normal range of similar working conditions, intercept the equipment operating temperature field distribution segment corresponding to this group of parameters and compare it with the temperature field segment of the normal state at the same stage to make a hotspot migration path comparison; According to the difference direction of hotspot migration, the judgment threshold of the parameter change order is modified until it conforms to the historical characteristics, and the modified threshold is associated with the temperature field characteristics and saved.

[0009] Furthermore, in the later stage of the conduction phase, it also includes: From each stage, we select the parameter interaction relationship that deviates most significantly from the historical pattern, and calculate the ratio of the instruction duration and response amplitude change corresponding to this relationship. If the ratio fluctuation exceeds the common range of similar working conditions, intercept the fluid pressure pulsation curve of the equipment operation at that stage and compare the pulsation period coincidence with the normal curve of the same state; According to the difference in period overlap, the analysis weight of the parameter interaction relationship is adjusted until it conforms to the historical characteristics, and the adjustment result is associated with the pressure pulsation characteristics and stored.

[0010] Furthermore, establishing the equipment benchmark response characteristics includes: grouping historical normal data according to the type of control instructions, extracting the response parameter change curve within the preset time window after each group of instructions is issued; fusing multiple curves of the same type of instructions to generate a benchmark template containing the response start threshold, steady-state fluctuation range, and peak occurrence time; establishing a benchmark feature sub-library for different operating conditions, which contains the typical time constant and parameter correlation coefficient of the instruction response under the condition.

[0011] Furthermore, the parameter micro-fluctuation transmission stages are divided into the following: calculating the second-order derivative of the deviation evolution trajectory, and taking the inflection point where the derivative changes from positive to negative as the stage node; the first stage records the propagation path of the initial fluctuation parameters, the second stage counts the number of parameters involved in the fluctuation and the intensity of their influence, and the third stage calculates the attenuation rate of the fluctuation amplitude of each parameter.

[0012] Furthermore, after outputting the dynamic evolution trajectory of the deviation feature, it also includes: Extract segments from the trajectory where the deviation increase exceeds the common value of similar working conditions within several consecutive sampling periods, and align the starting and ending points of the instruction execution segment and the response change segment corresponding to the segment; If the slope fluctuation amplitude of the response change section after alignment exceeds the historical normal range, intercept the equipment operation vibration waveform in this section and compare the peak interval with the earlier waveform of the same type of deviation; According to the difference in peak intervals, the judgment scale of the deviation increase is adjusted, and the characteristic fragments are re-extracted until they conform to the historical fluctuation pattern. The adjusted scale is associated with the vibration waveform characteristics and archived.

[0013] Furthermore, the process of identifying behavioral deviations also includes: Perform a superposition analysis on the parameter change curves of multiple consecutive behavioral deviations under the same instructions, and calculate the area ratio of the curve overlap area; If the proportion is lower than the set value, the interval with the most significant deviation in each curve is intercepted and the valley depth is compared with the parameter curve of the same historical degradation stage. According to the direction of valley depth difference, the identification threshold of behavioral deviation is corrected, and the analysis is re-superimposed until the overlapping area meets the standard. The corrected threshold is stored together with the curve characteristics.

[0014] In a second aspect, the present application provides a data-driven air compressor control system, the system comprising: Acquisition module, synchronously obtains the physical parameters of the air compressor and the device response data corresponding to each control instruction; Establish a module to establish the equipment baseline response characteristics corresponding to different control instructions based on historical normal operation data; The recognition module compares real-time response data with the baseline response characteristics to identify response drift, delay extension, feedback structure changes, and behavioral deviations under similar instructions when the control logic remains unchanged, and outputs the dynamic evolution trajectory of the deviation characteristics; When the behavior deviation exceeds the set range, the output module first divides the parameter micro-fluctuation transmission stage into key turning points of the deviation evolution trajectory. Then, the influence relationship between the parameters in each stage is verified to reproduce the transmission law of the historical degradation process under similar working conditions. If the change sequence and intensity distribution of the parameter influence relationship in each stage are consistent with the historical characteristics, even if the physical parameters are within the limit, it is still determined that the equipment has early signs of structural fatigue, control link abnormality, or internal medium state degradation. The feedback module adjusts the control instructions according to the risk determination results and feeds back the new instructions and response data to the expected response model unit to update the baseline characteristics.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention establishes a baseline response characteristic and compares it with the real-time response data to accurately identify the early response deviation of the equipment. It combines the conduction law of the historical degradation process to determine whether the equipment has early signs of failure, and adjusts the control instructions accordingly. This realizes intelligent and precise regulation of the air compressor, detects potential failures in advance, and ensures stable operation of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the air compressor.

[0017] Figure 2 It is a schematic diagram of the system framework structure of the present invention. DETAILED DESCRIPTION

[0018] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 2 This application provides a data-driven air compressor control method suitable for Figure 1 Air compressors, including: Synchronously obtain the physical parameters of the air compressor and the device response data corresponding to each control instruction; Based on historical normal operation data, establish the equipment benchmark response characteristics corresponding to different control instructions; Compare real-time response data with baseline response characteristics to identify response drift, delay extension, feedback structure changes, and behavioral deviations under similar instructions when the control logic remains unchanged, and output the dynamic evolution trajectory of the deviation characteristics; When the behavioral deviation exceeds the set range, the transmission stages of parameter micro-fluctuations are first divided based on the key turning points of the deviation evolution trajectory. The influence relationship between the parameters in each stage is then checked to see whether it reproduces the transmission pattern of the historical degradation process under similar working conditions. If the sequence of change and intensity distribution of the parameter influence relationship in each stage are consistent with the historical characteristics, even if the physical parameters are within the limit, it is still determined that the equipment has early signs of structural fatigue, control link abnormality, or internal medium degradation. The control instructions are adjusted according to the risk assessment results, and the new instructions and response data are fed back to the expected response model unit to update the baseline characteristics.

[0020] Among them, physical parameters refer to physical quantities that reflect the operating status of the air compressor, including temperature, pressure, flow, etc., which can be collected in real time through temperature sensors, pressure sensors, flow meters and other equipment to monitor the basic operating status of the air compressor.

[0021] Among them, equipment response data refers to the state change data presented by the air compressor after receiving the control command, including the speed change rate, pressure regulation rate, power consumption change, etc., which can be recorded in real time by the air compressor control system and used to analyze the equipment's response to the control command.

[0022] Among them, the benchmark response feature refers to the typical response pattern presented by the air compressor to different control instructions under the historical normal operating state. Specifically, it can be established by using the sliding window statistics method (the window size is set to 100 groups of data) to extract features (such as mean, variance, change trend, etc.) from the historical normal operating data, and used as a reference standard for judging whether the real-time response is normal.

[0023] The dynamic evolution trajectory of the deviation characteristic refers to the path along which the deviation generated by the real-time response data compared with the baseline response characteristic changes over time. This can be obtained by continuously recording the deviation values ​​and drawing a curve, which is used to intuitively display the development and changes of the equipment response deviation.

[0024] Among them, the conduction stage of parameter micro-fluctuations refers to dividing the transmission process of small parameter fluctuations between various equipment systems into different stages with the key turning points of the deviation evolution trajectory as the boundary. Each stage corresponds to a specific parameter influence relationship and change characteristics, which is used to deeply analyze the laws of deviation generation and transmission.

[0025] Among them, the conduction law of the historical degradation process refers to the law of the changing order and intensity distribution of the influence relationship between various parameters when the air compressor has suffered faults such as structural fatigue, control link abnormality or internal medium state degradation in the past. It can be obtained by analyzing and summarizing historical fault data to assist in judging whether the current equipment has signs of early faults.

[0026] Among them, the expected response model unit refers to the model unit used to store and update the benchmark response characteristics. The gradient descent method is used to continuously optimize the benchmark characteristics according to the new control instructions and response data to make them more consistent with the current actual operating status of the air compressor.

[0027] The innovation of this application lies in that by establishing a baseline response characteristic and comparing it with the real-time response data, the early response deviation of the equipment can be accurately identified. Combined with the conduction law of the historical degradation process, it is judged whether the equipment has early signs of failure, and the control instructions are adjusted accordingly, thereby realizing intelligent and precise regulation of the air compressor, discovering potential failures in advance, and ensuring stable operation of the equipment.

[0028] The working principle of this application is: first, the system synchronously collects the physical parameters of the air compressor and the equipment response data corresponding to each control instruction to provide basic data for subsequent analysis; secondly, the historical normal operation data is used to establish the equipment benchmark response characteristics corresponding to different control instructions as the basis for judging whether the equipment response is normal; then, the real-time response data is compared with the benchmark response characteristics to identify various response deviations, and the dynamic evolution trajectory of the deviation characteristics is output to promptly detect abnormal changes in the equipment response; then, when the behavioral deviation exceeds the set range, the conduction stage of the parameter micro-fluctuation is divided, and the influence relationship between the parameters in each stage is checked to see whether it conforms to the conduction law of the historical degradation process, so as to determine whether the equipment has early signs of failure, and even if the physical parameters are not out of limit, problems can be discovered in time; finally, the control instructions are adjusted according to the risk judgment results, and the new instructions and response data are fed back to the expected response model unit to update the benchmark characteristics so that the regulation is more adapted to the actual state of the equipment.

[0029] As a preferred embodiment, the data-driven air compressor control method of the present application is specifically implemented as follows: First, temperature sensors, pressure sensors, flow meters and other equipment are installed on the air compressor to collect physical parameters in real time, such as cylinder temperature (range 80-120°C), exhaust pressure (range 0.7-0.8MPa), exhaust flow (range 5-8m³ / min), etc. At the same time, the air compressor control system records equipment response data, such as speed change rate and pressure regulation rate.

[0030] Based on the past year's historical normal operation data, a sliding window statistical method (window size 100 groups) was used to establish baseline response characteristics for different control commands. For example, for the load command, the baseline response characteristics are: the time it takes for the pressure to increase from 0.4 MPa to 0.7 MPa is 15 ± 2 seconds, and the speed change rate is stable at around 5% / s.

[0031] During real-time operation, after the loading command was issued, the real-time response data showed that the time for the pressure to rise from 0.4MPa to 0.7MPa was 20s, and the speed change rate fluctuated greatly. After comparing with the benchmark response characteristics, it was identified that there was a prolonged response delay and behavioral deviation, and the dynamic evolution trajectory of the output deviation characteristics was obtained.

[0032] Because the behavioral deviation exceeded the set range (time deviation exceeded 5 seconds), the point in the trajectory where the pressure rise rate significantly slowed was used as the key turning point to divide the transmission stages of the parameter micro-fluctuation. Verification found that the order and intensity distribution of the influence relationships between parameters such as temperature, pressure, and speed in each stage were consistent with the transmission patterns of degradation processes caused by historical control link anomalies. Therefore, it was determined that the control link showed early signs of anomaly, even though the physical parameters were within the normal range at this time.

[0033] Based on the judgment results, the execution parameters of the loading instruction are adjusted, such as appropriately increasing the initial loading power. The new instruction and corresponding response data are fed back to the expected response model unit, and the baseline response characteristics corresponding to the loading instruction are updated through the gradient descent method, making subsequent control more precise.

[0034] This implementation method can help to discover early signs of air compressor failure in advance, adjust control instructions in time, effectively avoid further development of the failure, and improve the operating reliability and service life of the air compressor.

[0035] In some embodiments of the present application, after determining that the device has early signs of structural fatigue, control link abnormality, or internal medium state degradation, there is a problem of how to accurately locate the source of the deviation and refine the abnormality analysis to improve the reliability of the determination result.

[0036] In this regard, the present application further proposes to obtain the parameter influence relationship in each conduction stage that is inconsistent with the historical feature variation order, and mark it as an abnormal action point; the abnormal action point is used as the starting point to obtain the response time difference sequence of the parameter after the control instruction is issued, and compare it with the time difference sequence characteristics of the historical normal response; if the interval fluctuation amplitude of a certain link in the time difference sequence exceeds the normal range of the same working condition, the equipment operation soundprint or vibration spectrum fragment corresponding to the link is intercepted, and the resonance peak is compared with the soundprint / spectrum of the same link in the early stage of historical degradation; if the resonance peak offset does not reach the warning threshold, the analysis dimension of the parameter action relationship is adjusted and the variation order is rechecked; after each round of sorting, the number of abnormal action points is counted to see whether it has decreased. If it has not decreased for two consecutive times, the typical deviation conduction path of the same model equipment in the same degradation scenario is introduced as a reference, and the current time difference sequence is checked for consistency with the reference path link by link; until the number of abnormal action points is lower than the set threshold, the final located deviation source link and verification parameters are associated with the early sign judgment result, and updated to the benchmark feature library of the expected response model unit.

[0037] Specifically, after identifying early signs of equipment degradation, the system first identifies abnormal points that are inconsistent with the historical characteristic progression. Using these points as a starting point, it tracks the response time difference sequence and identifies the abnormal fluctuations by comparing them with the characteristics of normal time differences. This system then compares the resonance peaks of the voiceprint or vibration spectrum to verify, at the physical signal level, whether the deviation has reached a warning level. If the threshold is not reached, the system adjusts the analysis dimensions and re-analyzes the analysis to reduce the number of abnormal points. If the number of abnormal points is difficult to reduce, the system uses typical degradation paths of similar equipment as a reference to precisely locate the source of the deviation. This multi-step, multi-dimensional analysis approach penetrates complex parameter correlations to accurately identify the root cause of the problem, making subsequent control command adjustments more targeted. This method is particularly effective in the early stages of equipment degradation, when parameter fluctuations are subtle and complex, effectively capturing key abnormalities and preventing potential risks from being overlooked early, thereby improving the safety of air compressor operation and the effectiveness of control. Simultaneously, the source location results are updated to the baseline feature library, enabling the model to continuously learn new abnormal patterns, enhancing the system's self-optimization capabilities and long-term adaptability.

[0038] In the technical solution of this application, the identification and marking of abnormal action points can be achieved in a variety of ways. For example, an association rule mining algorithm can be used to identify nodes that are inconsistent with the historical change order from the parameter influence relationship. When obtaining and comparing the response time difference sequence, time series analysis methods such as dynamic time warping (DTW) can be used to measure the similarity between the real-time time difference sequence and the historical normal sequence.

[0039] Resonance peak comparison can be performed using spectrum analysis tools such as the Fast Fourier Transform (FFT) to convert voiceprint or vibration signals into a spectrum. Peak detection algorithms can then be used to extract the resonance peaks for comparison. When adjusting the analysis dimension of parameter interactions, dimensionality reduction algorithms such as Principal Component Analysis (PCA) can be used to focus on key parameters. By using typical deviation transmission paths from the same device model as a reference, a path matching model can be established, such as using graph theory-based path similarity calculation or using deep learning-based graph neural networks (GNNs) to learn path features and perform matching.

[0040] The technical solution of this application addresses the problem of accurately analyzing early signs of air compressor degradation by first identifying anomaly points and pinpointing parameter associations that could indicate potential issues. This step effectively filters out key nodes that deviate from historical patterns within a complex parameter transmission network. For example, when an air compressor transitions from normal operation to a stage of mild degradation, certain parameters (such as the correlation between air pressure regulation response time and motor speed) may deviate from the historically normal progression. Association rule mining can quickly identify these anomaly points.

[0041] Track the response time difference sequence starting from the abnormal point of action. For example, the time difference sequence of the oil-gas separator pressure rise after the control command is issued is compared with the historical normal sequence. If the fluctuation amplitude of a certain link (such as the pressure sensor feedback delay) is found to be outside the range of similar operating conditions, the operating soundprint (such as the separator housing vibration) or vibration spectrum of this link is intercepted. After FFT conversion, it is compared with the spectrum of the same link at the beginning of the historical degradation period. If the resonance peak shift is small (not reaching the warning threshold), PCA is used to adjust the analysis dimension, focusing on the core parameters related to pressure feedback (such as sensor voltage and pipeline resistance), and re-examining the parameter transition sequence.

[0042] If the number of anomaly points persists after two rounds of analysis (e.g., three anomalies remain), a typical deviation transmission path for the same model of equipment under an oil-gas separator degradation scenario (e.g., "motor load fluctuation -- separator pressure anomaly -- sensor feedback delay") is introduced. The consistency of the current time difference sequence with this path is then verified step by step. Assuming an 80% match between the current sequence and the typical path, the source of the deviation can be identified as a clogged filter element within the separator. This result and corresponding verification parameters (e.g., pressure differential across the filter element and vibration spectrum peak) are then correlated with the early sign determination results and updated to the baseline feature library.

[0043] Through this approach, the air compressor control system can accurately locate the root cause of the problem in the early stages of degradation, making the adjustment of control instructions more targeted (such as appropriately increasing the frequency of filter element cleanliness monitoring). At the same time, the updated baseline feature library can more accurately identify similar signs of degradation, significantly improving the equipment's operational reliability and intelligent control level.

[0044] In some embodiments of the present application, in the early stage of dividing the parameter micro-fluctuation conduction stage, when the parameter combination does not match the intensity distribution of the historical degradation conduction law, there is still a challenge in how to effectively calibrate the accuracy of parameter analysis and reduce misjudgments caused by data sampling or waveform feature deviations.

[0045] In this regard, the present application further proposes to obtain parameter combinations that do not match the intensity distribution of the historical degradation conduction law in each stage, and align the control instruction curves corresponding to these parameters with the equipment response curves to perform peak alignment; if the number of curve intersections exceeds the preset range of similar working conditions in multiple consecutive fluctuation cycles after alignment, intercept the equipment operation vibration waveform in the intersection-dense interval and compare it with the vibration waveform in the normal state of the same stage to perform trough spacing comparison; according to the direction of the trough spacing difference, adjust the sampling interval of the parameter combination, recheck the intensity distribution until it meets the historical characteristics, and archive the adjusted sampling interval and vibration waveform characteristics together.

[0046] Specifically, during the initial phase of parameter micro-fluctuation transmission, if the intensity distribution of a parameter combination (e.g., compressor intake pressure and exhaust flow) is found to be inconsistent with historical patterns, the system first performs peak alignment on the corresponding control command curve (e.g., intake valve control signal) and response curve (e.g., exhaust flow feedback signal) to eliminate phase deviation on the time axis. If, after alignment, the number of intersections between the two curves exceeds a preset threshold for similar operating conditions (e.g., normally ≤ 2 intersections, currently 5), the system then intercepts the vibration waveform (e.g., compressor cylinder vibration signal) in the region with the highest number of intersections and compares the trough spacing with the vibration waveform under normal conditions at the same stage. For example, the trough spacing of the normal waveform is stable at 0.2 seconds, while the trough spacing of the current waveform fluctuates wildly between 0.15 and 0.25 seconds. Based on the direction of the trough spacing discrepancy (e.g., overall smaller), the sampling interval for that parameter combination is adjusted (e.g., from 10ms to 8ms), and the intensity distribution is recalculated until it matches the historical patterns. The 8ms sampling interval and the fluctuation waveform characteristics are archived.

[0047] This method reverse-calibrates the sampling interval through precise comparison of waveform features, effectively resolving the problem of intensity distribution misjudgment caused by inappropriate data acquisition frequency. Especially in the early stages of micro-fluctuations in equipment parameters, when signal characteristics are weak and susceptible to noise, this method can identify critical deviations through the geometric characteristics of peaks and troughs, ensuring accurate delineation of the conduction phase. Furthermore, the archived sampling intervals and waveform features can serve as analysis templates for similar operating conditions, reducing the cost of repeated calibration, improving the system's sensitivity and response speed to parameter fluctuations, and laying a data foundation for accurately identifying early signs of equipment degradation.

[0048] In the technical solution of this application, peak alignment of the control command curve and the device response curve can be achieved in a variety of ways. For example, a peak detection algorithm (such as threshold-based local maximum identification) can be used to locate the peak positions of the two curves, and then the time axis alignment can be achieved by translating or stretching the curves.

[0049] The comparison of the trough spacing of the vibration waveform can be assisted by signal processing tools, such as extracting the low-frequency components of the waveform through wavelet transform to highlight the trough characteristics, and then using Euclidean distance or Manhattan distance to calculate the spacing difference with the normal waveform; the adjustment of the parameter combination sampling interval can be based on the statistical characteristics of the trough spacing difference. For example, when the difference is systematically smaller, the sampling interval is shortened proportionally (for example, if the difference amplitude is 20%, the sampling interval is shortened by 20%).

[0050] The technical solution of this application addresses the accuracy issue of dividing the parameter micro-fluctuation conduction phase by first eliminating the time phase difference between the control command and the response curve through peak alignment, ensuring the validity of the basic data for parameter correlation analysis. For example, when the air compressor is in a low-load condition, there is a 0.5-second delay between the peak of the intake valve control command (maximum opening) and the peak of the exhaust flow response. If the analysis is conducted directly without alignment, the intensity distribution will show "weak correlation between command and response", which does not match the historical pattern (strong correlation). After locating the peak through peak detection, the response curve is shifted by 0.5 seconds to achieve alignment. At this time, the recalculated number of intersections can truly reflect the dynamic correlation between the two.

[0051] If the number of intersections remains abnormal after alignment (e.g., outside the preset range), the physical deviation is identified by measuring the trough spacing of the vibration waveform. For example, a certain parameter combination exhibited a high number of intersections after alignment. The vibration waveform of an air compressor crankcase was intercepted and found to be stable at a 0.3-second trough spacing under normal conditions. However, due to slight bearing wear, the trough spacing in the current waveform was shortened to 0.25 seconds and fluctuated frequently. Based on this discrepancy (small spacing), the sampling interval for this parameter combination was adjusted from 15ms to 12ms (a proportional 20% reduction). After re-collecting the data, the matching degree between the intensity distribution and the historical degradation pattern increased from 60% to 92%. The vibration waveform characteristics at the 12ms sampling interval and the initial stage of wear were archived.

[0052] This method calibrates data collection standards early in the process of parameter micro-fluctuations, preventing misjudgments of patterns due to inappropriate sampling intervals and ensuring the accuracy of subsequent early sign determinations. Furthermore, archived feature data can be used to train a predictive model for parameter sampling intervals. When similar operating conditions recur, the optimal sampling interval is automatically invoked, significantly improving the system's analysis efficiency and response speed, providing a more reliable foundation for refined air compressor control and fault warning.

[0053] The present application further proposes to find the group of parameter combinations in each stage that deviates the most from the historical transition sequence, and to count the interval values ​​between the instruction triggering moment and the response extreme value appearance moment corresponding to this group; if the discreteness of the interval value exceeds the normal range of the same working condition, intercept the equipment operating temperature field distribution fragment corresponding to this group of parameters, and compare the hotspot migration path with the temperature field fragment of the normal state in the same stage; according to the direction of the hotspot migration difference, correct the judgment threshold of the parameter transition sequence until it meets the historical characteristics, and associate the corrected threshold with the temperature field characteristics and save it.

[0054] Specifically, when the deviation in the order of change of a parameter group (such as the flow rate and outlet temperature of the cooling system of an air compressor) is the largest, the dispersion of the time interval between the command trigger and the response extreme value is first calculated. If the dispersion is too large (for example, the standard deviation exceeds 0.3 seconds), the temperature field distribution of the corresponding time period is retrieved (such as infrared thermal imaging of the radiator area) and compared with the hotspot migration path under normal conditions (for example, hotspots normally spread from the center to the edge, but currently they are locally concentrated). Based on the direction of the migration difference (for example, the aggregation speed is faster than historical data), the threshold for determining the order of change of the parameter group is revised (for example, from the original 1.2 seconds to 1.0 seconds) until it matches the historical characteristics, and the threshold is associated with the temperature field aggregation characteristics and archived.

[0055] This approach leverages the physical characteristics of the temperature field to provide a basis for correction when parameter-time correlations are ambiguous, effectively eliminating the limitations of single-time-dimensional analysis. This approach is particularly effective in scenarios where parameter interactions intensify during mid-stage equipment degradation, accurately identifying the criteria for determining the order of change and providing a reliable foundation for subsequent risk assessment. Furthermore, the use of associated archived features can accelerate the resolution of similar issues, forming a closed-loop optimization mechanism.

[0056] At the implementation level, the dispersion of interval values ​​can be calculated using the coefficient of variation; hotspot migration in the temperature field can be compared using an image analysis algorithm based on contour matching; and threshold correction can be achieved by establishing a mapping model between hotspot migration speed and threshold. Each step is logically progressive: the dispersion of interval values ​​triggers temperature field analysis, hotspot migration differences determine the direction of threshold correction, and finally, the archived results feed back into the verification process, improving overall analysis accuracy.

[0057] In some embodiments of the present application, when dividing the parameter micro-fluctuation conduction stage, the accuracy of parameter analysis has been gradually improved through the early sampling interval calibration and the mid-term judgment threshold correction. However, in the later stage of the conduction stage, when the parameter interaction relationship deviates significantly from the historical rules, how to optimize the weight distribution of the interaction relationship through an analysis dimension that is more in line with the physical characteristics of the device still remains a challenge.

[0058] In this regard, the present application further proposes to select the parameter interaction relationship with the most significant deviation from the historical rules from each stage, and to count the ratio of the instruction duration and the response amplitude change corresponding to the relationship; if the ratio fluctuation exceeds the common range of similar working conditions, the fluid pressure pulsation curve of the equipment operation in this stage is intercepted, and the pulsation cycle overlap is compared with the normal curve of the same state; according to the difference in cycle overlap, the analysis weight of the parameter interaction relationship is adjusted until it conforms to the historical characteristics, and the adjustment result is associated with the pressure pulsation characteristics and stored.

[0059] Specifically, in the later stage of the divided conduction stage, when the interaction relationship between a certain parameter (such as the intake valve opening of the air compressor and the pressure in the cylinder) deviates most significantly from the historical law, the system first calculates the ratio of the instruction duration (such as the valve opening for 1.5 seconds) and the response amplitude change (such as the pressure in the cylinder rises by 0.3MPa) under this relationship. If the fluctuation range of this ratio (such as the normal range is 0.2-0.25MPa / second, currently 0.15-0.3MPa / second) exceeds the common range, the fluid pressure pulsation curve of the corresponding stage (such as the high-frequency pulsation signal collected by the cylinder pressure sensor) is intercepted and compared with the normal curve in the same state for the coincidence of the pulsation period - for example, the pulsation period of the normal curve is stable at 0.02 seconds, and the coincidence is 90%, while the current curve has periodic fluctuations due to valve leakage, and the coincidence is reduced to 60%. Based on the difference in period overlap (reduced by 30%), the analysis weight of the parameter interaction relationship is adjusted (for example, from the original 0.3 to 0.5). After recalculation, the interaction relationship is made consistent with historical characteristics, and the adjusted weight is associated and stored with the pulsation period disorder characteristics caused by the leakage.

[0060] This method leverages the periodic characteristics of fluid pressure pulsation to inversely optimize the weights of parameter interactions, effectively addressing parameter correlation deviations caused by fluid dynamic anomalies such as valve wear and pipeline blockage. Particularly in the late conduction phase, when parameter interactions are exacerbated by physical degradation, this method uses pressure pulsation, a signal that directly reflects the internal state of the device, to precisely calibrate the analysis weights, ensuring that the final conduction phase classification results accurately reflect the actual degradation process.

[0061] In the technical solution of the present application, the ratio calculation of the instruction duration and the response amplitude change can be achieved through the sliding window method, and the fluctuation range of the ratio can be tracked in real time; the comparison of the fluid pressure pulsation period overlap can be achieved by using spectrum analysis combined with dynamic time warping (DTW) algorithm, first extracting the periodic features through Fourier transform, and then calculating the similarity with the normal curve; the adjustment of the parameter interaction relationship analysis weight can be based on the mapping model of the period overlap difference and the weight, such as using proportional adjustment (for every 10% decrease in overlap, the weight increases by 0.1).

[0062] In some embodiments of this application, including some of the aforementioned embodiments, a fundamental approach to establishing a baseline device response signature is proposed as a reference for identifying parameter deviations and determining device status. However, there is still room for improvement in how to make the baseline signature more tailored to the actual characteristics of different control instruction types and operating conditions, and to avoid insufficient recognition accuracy due to the overly universal nature of the baseline template.

[0063] In this regard, the present application further proposes to group historical normal data according to the type of control instructions, extract the response parameter change curve within the preset time window after each group of instructions is issued; fuse multiple curves of the same type of instructions to generate a benchmark template containing the response start threshold, steady-state fluctuation range, and peak occurrence time; establish a benchmark feature sub-library for different operating conditions, which contains the typical time constants and parameter correlation coefficients of the instruction response under the operating condition.

[0064] Specifically, to establish a baseline response signature, the system first groups historical normal data by control command type (e.g., "load command," "unload command," and "pressure adjustment command" for an air compressor). For each command group, the system extracts the response parameter change curve (e.g., the time-dependent changes in parameters such as pressure, temperature, and speed) within 5 seconds of issuance. For example, the system fuses the 100 curves in the "load command" group (by calculating the mean and standard deviation of each curve at the same time point) to generate a baseline template for that command: the response startup threshold (e.g., a pressure increase of 0.1 MPa is considered startup), the steady-state fluctuation range (e.g., pressure within 0.8 ± 0.02 MPa after stabilization), and the peak occurrence time (e.g., peak pressure reaches 2.3 seconds after loading). Furthermore, sub-libraries are created for each operating condition (e.g., light load, full load, low temperature environment, and high temperature environment). Each sub-library contains the typical time constant for that operating condition (e.g., the time constant for pressure to reach steady state at full load is 1.8 seconds) and parameter correlation coefficients (e.g., the correlation coefficient between load current and exhaust pressure is 0.92).

[0065] This approach, through two-tiered segmentation, elevates the benchmark signature from a single template to a multi-dimensional, scenario-based reference system. This approach is particularly effective in scenarios where equipment response patterns vary significantly with commands and operating conditions (for example, the pressure rise rate of an air compressor loading command under full and light loads can differ by up to 30%). This approach effectively avoids misjudgments of normal deviations and missed detections of abnormal deviations due to generalization of the benchmark.

[0066] In the technical solution of this application, response parameter change curves can be extracted using a sliding time window, with the window size set according to the instruction execution cycle (e.g., 3 seconds for short instructions and 10 seconds for long instructions). Curve fusion can be achieved using statistical methods (e.g., ±3 times the standard deviation of the mean curve) or clustering algorithms (e.g., K-means extraction of typical curves). The characteristics of the reference template (startup threshold, peak time, etc.) can be automatically identified using a feature point detection algorithm. The operating condition sub-library can be divided based on the clustering results of operating condition parameters (e.g., load factor, ambient temperature, and intake pressure). The time constant can be calculated by fitting the response curve with a first-order inertia model, and the parameter correlation coefficient can be analyzed using the Pearson correlation coefficient or partial correlation coefficient.

[0067] In some embodiments of this application, the classification of parameter micro-fluctuation transmission stages is a key step in analyzing device state changes and identifying early signs of degradation. However, there is a challenge in extracting the core features of each stage to avoid overly general stage classification or lack of targeted feature extraction.

[0068] In this regard, the present application further proposes to calculate the second-order derivative of the deviation evolution trajectory, and take the inflection point where the derivative changes from positive to negative as the stage node; the first stage records the propagation path of the initial fluctuation parameters, the second stage counts the number of parameters involved in the fluctuation and the intensity of the influence, and the third stage calculates the attenuation rate of the fluctuation amplitude of each parameter.

[0069] Specifically, when dividing the parameter micro-fluctuation transmission stage, the system first calculates the second-order derivative of the deviation evolution trajectory (such as the curve of the air compressor pressure deviation change over time). The second-order derivative reflects the increase or decrease trend of the deviation change rate. When the derivative changes from positive to negative, it indicates that the growth rate of the deviation fluctuation has begun to slow down. The inflection point at this time is used as the stage node. For example, the second-order derivative of a certain deviation trajectory has an inflection point from positive to negative at 10 seconds and 25 seconds. Therefore, the transmission process is divided into three stages: 0-10 seconds (first stage), 10-25 seconds (second stage), and after 25 seconds (third stage); The first stage focuses on the source and propagation of the initial fluctuation, recording how the first parameter that fluctuates (such as intake pressure) affects related parameters (such as flow rate and temperature), and forming a propagation path map (such as "intake pressure-flow rate-cylinder temperature"); the second stage focuses on analyzing the degree of diffusion of the fluctuation, counting the number of newly added parameters participating in the fluctuation (such as from 2 to 5), and quantifying the role of each parameter through the impact intensity coefficient (such as the proportion of the increase in the deviation of other parameters caused by the fluctuation of a certain parameter); the third stage focuses on the convergence trend of the fluctuation, and calculates the decay rate of the fluctuation amplitude of each parameter over time (such as the rate at which the pressure fluctuation amplitude decreases from 0.2MPa to 0.05MPa is 0.03MPa / second).

[0070] This method divides parameter fluctuations into stages based on mathematical characteristics (second-order derivative inflection points), closely linking stage nodes to the inherent dynamics of the fluctuations rather than simply segmenting them over time. Feature extraction at each stage focuses on its own characteristics, fully capturing the entire process of microfluctuations from their origin to their intensified diffusion and gradual convergence. This method is particularly effective in scenarios involving early stages of device degradation, when parameter fluctuations are subtle and transmission pathways are complex. This method accurately captures the stage-specific characteristics of the fluctuations, providing a clear analytical framework for identifying anomalous transmission patterns.

[0071] In the technical solution of the present application, the calculation of the second-order derivative of the deviation evolution trajectory can be achieved through a numerical differentiation algorithm (such as the finite difference method), and the inflection point identification can be combined with threshold judgment (the inflection point is confirmed when the second-order derivative changes from positive to negative for three consecutive sampling points); the propagation path record in the first stage can adopt a directed graph model, with nodes representing parameters and directed edges representing influence relationships; the influence intensity statistics in the second stage can calculate the degree of correlation between parameters through the Pearson correlation coefficient or mutual information entropy; the decay rate calculation in the third stage can be achieved by linearly fitting the curve of the fluctuation amplitude changing with time, and the slope is the decay rate.

[0072] In some embodiments of this application, the dynamic evolution trajectory of output deviation features provides fundamental data for identifying equipment anomalies. However, segments of the trajectory with abnormal deviation increases often contain critical early degradation information. Accurately extracting these segments and calibrating the deviation increase determination criteria to avoid feature omissions or misjudgments due to inappropriate determination scales presents challenges.

[0073] In this regard, the present application further proposes to extract from the trajectory a segment in which the deviation increase exceeds the common value of the same working condition within several consecutive sampling periods, and align the start and end points of the instruction execution segment and the response change segment corresponding to the segment; if the slope fluctuation amplitude of the response change segment exceeds the historical normal range after alignment, intercept the equipment operation vibration waveform in this interval and compare it with the peak interval of the early waveform of the same type of deviation; according to the difference in peak interval, adjust the judgment scale of the deviation increase, re-extract the feature segment until it conforms to the historical fluctuation law, and associate the adjusted scale with the vibration waveform feature for archiving.

[0074] Specifically, after outputting the dynamic evolution trajectory, the system first selects segments from the trajectory where the deviation increase (e.g., an increase in compressor pressure deviation from 0.05 MPa to 0.15 MPa) exceeds the typical value for similar operating conditions (e.g., a typical increase ≤ 0.08 MPa) within three consecutive sampling periods. The system then aligns the corresponding command execution segment (e.g., the 0-2 seconds of a "pressure increase" command) with the response change segment (e.g., the 1-3 seconds of a pressure increase from 0.8 MPa to 0.95 MPa) to eliminate time offset. If the slope fluctuation amplitude of the response change segment after alignment (e.g., a normal slope of 0.07 MPa / second, but currently fluctuating wildly between 0.05 and 0.09 MPa / second) exceeds the historical range, the system then intercepts the corresponding vibration waveform (e.g., the cylinder vibration signal) and compares the peak-to-peak interval with the normal waveform from an earlier period of the same deviation. For example, the normal waveform peak-to-peak interval is 0.1 second, but now it has dropped to 0.08-0.12 seconds due to loose components. Based on the difference in peak intervals (a 40% increase in fluctuation amplitude), the judgment scale for the deviation increase is adjusted (for example, the common value is relaxed from 0.08MPa to 0.1MPa). After re-extracting the fragments, they are made to conform to historical rules, and the adjusted scale is associated with the vibration waveform of the loosening feature and archived.

[0075] This method uses the physical characteristics of the vibration waveform to reverse-calibrate the deviation amplitude determination criteria, effectively addressing complex scenarios that are difficult to address with single numerical analysis. Especially when the deviation amplitude is near a critical value, the difference in the vibration peak interval provides a more sensitive basis for determination, helping to avoid missed anomalies due to rigid scaling.

[0076] In the technical solution of this application, the alignment of the instruction execution segment and the response change segment can be achieved using a dynamic time warping (DTW) algorithm, allowing for a certain amount of time distortion to achieve optimal matching. The slope fluctuation amplitude of the response change segment can be determined by calculating the standard deviation of the slopes of adjacent sampling points. The peak interval comparison of the vibration waveform can be combined with peak detection and sliding window statistics to quantify the mean and fluctuation range of the interval. The adjustment of the judgment scale can establish a mapping relationship between the peak interval difference and the scale correction amount (for example, for every 10% increase in interval fluctuation, the scale is relaxed by 5%) to ensure that the adjusted scale is consistent with the physical signal characteristics.

[0077] In some embodiments of the present application, identifying behavioral deviations under similar instructions is an important step in determining whether there is an abnormality in the device. However, when the consistency of the parameter change curves of multiple consecutive deviations is low, it is difficult to accurately define the deviation judgment criteria through only a single curve analysis, which may lead to insufficient reliability of the recognition results.

[0078] In this regard, the present application further proposes to perform a superposition analysis on the parameter change curves of multiple consecutive behavioral deviations under similar instructions, and to calculate the area ratio of the curve overlap area; if the ratio is lower than the set value, the interval with the most significant deviation in each curve is intercepted, and the valley depth is compared with the parameter curve of the historical similar early degradation stage; according to the direction of the valley depth difference, the recognition threshold of the behavioral deviation is corrected, and the superposition analysis is performed again until the overlap area meets the standard, and the corrected threshold is stored together with the curve characteristics.

[0079] Specifically, when identifying behavioral deviations, the system first overlays parameter change curves (e.g., pressure drop curves) from five consecutive behavioral deviations under similar commands (e.g., an air compressor "unload" command) and calculates the percentage of overlap (e.g., ≥60% should be the normal range, but currently only 45%). If this percentage falls below a set value, the system then extracts the most significant deviation from each curve (e.g., the abnormal pressure drop) and compares the valley depth with historical parameter curves from similar early stages of degradation (e.g., valve leakage). For example, if the valley depth of the historical curve was 0.3 MPa, the valley depths of the five current curves ranged from 0.2 to 0.4 MPa. Based on the direction of the depth difference (overall shallower by 10%), the system then adjusts the threshold for identifying behavioral deviations (e.g., lowering the pressure deviation threshold from 0.2 MPa to 0.18 MPa). After re-overlaying and analyzing, the overlap percentage is increased to 65%. The adjusted threshold is then stored along with the valley characteristics of the batch of curves.

[0080] This method, through the identification and comparison of multiple curves, helps to resolve the randomness problem of single curve analysis. Especially in the early stages of degradation, when parameter deviation characteristics have not yet stabilized, it can correct the convergence and dispersion of curve characteristics through threshold correction and capture hidden common deviation patterns.

[0081] In the technical solution of the present application, the curve overlay analysis can use the point-by-point weighted averaging method to generate an average curve, and quantify the consistency by calculating the proportion of the overlapping area of ​​each curve with the average curve; the valley depth comparison can be combined with the extreme value detection algorithm (such as local minimum value identification) to extract the depth parameters, and use variance analysis to measure the difference with the historical curve; the correction of the recognition threshold can establish a linear mapping relationship between the valley depth difference and the threshold adjustment amount (such as the depth is 10% shallower, and the threshold is lowered by 8%) to ensure that the corrected threshold matches the actual deviation feature.

[0082] In a second aspect, the present invention further proposes a data-driven air compressor control system, which includes the following steps: Acquisition module, synchronously obtains the physical parameters of the air compressor and the device response data corresponding to each control instruction; Establish a module to establish the equipment baseline response characteristics corresponding to different control instructions based on historical normal operation data; The recognition module compares real-time response data with the baseline response characteristics to identify response drift, delay extension, feedback structure changes, and behavioral deviations under similar instructions when the control logic remains unchanged, and outputs the dynamic evolution trajectory of the deviation characteristics; When the behavior deviation exceeds the set range, the output module first divides the parameter micro-fluctuation transmission stage into key turning points of the deviation evolution trajectory. Then, the influence relationship between the parameters in each stage is verified to reproduce the transmission law of the historical degradation process under similar working conditions. If the change sequence and intensity distribution of the parameter influence relationship in each stage are consistent with the historical characteristics, even if the physical parameters are within the limit, it is still determined that the equipment has early signs of structural fatigue, control link abnormality, or internal medium state degradation. The feedback module adjusts the control instructions according to the risk determination results and feeds back the new instructions and response data to the expected response model unit to update the baseline characteristics.

[0083] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. The air compressor control method based on data drive is characterized by: include: Synchronously obtain the physical parameters of the air compressor and the device response data corresponding to each control instruction; Based on historical normal operation data, establish the equipment benchmark response characteristics corresponding to different control instructions; Compare real-time response data with baseline response characteristics to identify response drift, delay extension, feedback structure changes, and behavioral deviations under similar instructions when the control logic remains unchanged, and output the dynamic evolution trajectory of the deviation characteristics; When the behavioral deviation exceeds the set range, the transmission stages of parameter micro-fluctuations are first divided based on the key turning points of the deviation evolution trajectory. The influence relationship between the parameters in each stage is then checked to see whether it reproduces the transmission pattern of the historical degradation process under similar working conditions. If the sequence of change and intensity distribution of the parameter influence relationship in each stage are consistent with the historical characteristics, even if the physical parameters are within the limit, it is still determined that the equipment has early signs of structural fatigue, control link abnormality, or internal medium degradation. The control instructions are adjusted according to the risk assessment results, and the new instructions and response data are fed back to the expected response model unit to update the baseline characteristics.

2. The data-driven air compressor control method according to claim 1, characterized in that: After determining that the device has early signs, it also includes: Obtain the parameter influence relationship in each conduction stage that is inconsistent with the historical characteristic change order and mark it as an abnormal action point; Taking the abnormal action point as the starting point, obtain the response time difference sequence of the parameter after the control command is issued, and compare it with the characteristics of the time difference sequence of historical normal response; If the interval fluctuation amplitude of a certain link in the time difference sequence exceeds the normal range of similar working conditions, intercept the equipment operation soundprint or vibration spectrum fragment corresponding to this link and compare the resonance peak with the soundprint / spectrum of the same link at the early stage of historical degradation; If the resonance peak shift does not reach the warning threshold, adjust the analysis dimension of the parameter interaction relationship and recheck the transition order; After each round of sorting, count the number of abnormal action points to see if they have decreased. If they have not decreased for two consecutive times, use the typical deviation transmission path of the same model of equipment in the same degradation scenario as a reference and check the consistency of the current time difference sequence with the reference path step by step. Until the number of abnormal action points is lower than the set threshold, the final located deviation source link and verification parameters are associated with the early sign judgment results and updated to the baseline feature library of the expected response model unit.

3. The data-driven air compressor control method according to claim 1, characterized in that: In the early stage of the micro-fluctuation transmission phase of the division parameters, it also includes: Obtain parameter combinations that do not match the intensity distribution of the historical degradation conduction law in each stage, and align the control command curves corresponding to these parameters with the equipment response curves; If the number of curve intersections exceeds the preset range for similar operating conditions in multiple consecutive fluctuation cycles after alignment, intercept the equipment operating vibration waveform in the area with dense intersections and compare it with the vibration waveform in the normal state at the same stage for trough spacing; According to the difference in trough spacing, adjust the sampling interval of the parameter combination and recheck the intensity distribution until it conforms to the historical characteristics. Archive the adjusted sampling interval together with the vibration waveform characteristics.

4. The data-driven air compressor control method according to claim 1, characterized in that: In the mid-term of verifying the impact of parameters on the relationship, it also includes: Find the parameter combination with the largest deviation from the historical change order in each stage, and calculate the interval between the instruction triggering time and the response extreme value appearance time corresponding to this group; If the dispersion of the interval value exceeds the normal range of similar working conditions, intercept the equipment operating temperature field distribution segment corresponding to this group of parameters and compare it with the temperature field segment of the normal state at the same stage to make a hotspot migration path comparison; According to the difference direction of hotspot migration, the judgment threshold of the parameter change order is modified until it conforms to the historical characteristics, and the modified threshold is associated with the temperature field characteristics and saved.

5. The data-driven air compressor control method according to claim 1, characterized in that: In the later stages of the division and conduction phase, it also includes: From each stage, we select the parameter interaction relationship that deviates most significantly from the historical pattern, and calculate the ratio of the instruction duration and response amplitude change corresponding to this relationship. If the ratio fluctuation exceeds the common range of similar working conditions, intercept the fluid pressure pulsation curve of the equipment operation at that stage and compare the pulsation period coincidence with the normal curve of the same state; According to the difference in period overlap, the analysis weight of the parameter interaction relationship is adjusted until it conforms to the historical characteristics, and the adjustment result is associated with the pressure pulsation characteristics and stored.

6. The data-driven air compressor control method according to claim 1, characterized in that: Establishing a baseline device response characteristic involves: Group historical normal data by control instruction type and extract the response parameter change curve within the preset time window after each group of instructions is issued; Fusion of multiple curves of the same type of instructions to generate a benchmark template containing the response start threshold, steady-state fluctuation range, and peak occurrence time; A benchmark feature sub-library is established for each operating condition, which contains the typical time constants and parameter correlation coefficients of the command response under the condition.

7. The data-driven air compressor control method according to claim 1, characterized in that: The parameter micro-fluctuation conduction stage is divided into: Calculate the second-order derivative of the deviation evolution trajectory, and take the inflection point where the derivative changes from positive to negative as the stage node; The first stage records the propagation path of the initial fluctuation parameters, the second stage counts the number of parameters involved in the fluctuation and the intensity of their influence, and the third stage calculates the attenuation rate of the fluctuation amplitude of each parameter.

8. The data-driven air compressor control method according to claim 1, characterized in that: After outputting the dynamic evolution trajectory of the deviation feature, it also includes: Extract segments from the trajectory where the deviation increase exceeds the common value of similar working conditions within several consecutive sampling periods, and align the starting and ending points of the instruction execution segment and the response change segment corresponding to the segment; If the slope fluctuation amplitude of the response change section after alignment exceeds the historical normal range, intercept the equipment operation vibration waveform in this section and compare the peak interval with the earlier waveform of the same type of deviation; According to the difference in peak intervals, the judgment scale of the deviation increase is adjusted, and the characteristic fragments are re-extracted until they conform to the historical fluctuation pattern. The adjusted scale is associated with the vibration waveform characteristics and archived.

9. The data-driven air compressor control method according to claim 1, characterized in that: The process of identifying behavioral deviations also includes: Perform a superposition analysis on the parameter change curves of multiple consecutive behavioral deviations under the same instructions, and calculate the area ratio of the curve overlap area; If the proportion is lower than the set value, the interval with the most significant deviation in each curve is intercepted and the valley depth is compared with the parameter curve of the same historical degradation stage. According to the direction of valley depth difference, the identification threshold of behavioral deviation is corrected, and the analysis is re-superimposed until the overlapping area meets the standard. The corrected threshold is stored together with the curve characteristics.

10. A data-driven air compressor control system, applicable to the data-driven air compressor control method according to any one of claims 1 to 9, characterized in that: The system comprises: Acquisition module, synchronously obtains the physical parameters of the air compressor and the device response data corresponding to each control instruction; Establish a module to establish the equipment baseline response characteristics corresponding to different control instructions based on historical normal operation data; The recognition module compares real-time response data with the baseline response characteristics to identify response drift, delay extension, feedback structure changes, and behavioral deviations under similar instructions when the control logic remains unchanged, and outputs the dynamic evolution trajectory of the deviation characteristics; When the behavior deviation exceeds the set range, the output module first divides the parameter micro-fluctuation transmission stage into key turning points of the deviation evolution trajectory. Then, the influence relationship between the parameters in each stage is verified to reproduce the transmission law of the historical degradation process under similar working conditions. If the change sequence and intensity distribution of the parameter influence relationship in each stage are consistent with the historical characteristics, even if the physical parameters are within the limit, it is still determined that the equipment has early signs of structural fatigue, control link abnormality, or internal medium state degradation. The feedback module adjusts the control instructions according to the risk determination results and feeds back the new instructions and response data to the expected response model unit to update the baseline characteristics.

Images