Big data analysis and treatment system and method based on artificial intelligence

Abstract

The invention belongs to the technical field of data analysis, and particularly discloses a big data analysis and treatment system and method based on artificial intelligence, and the system comprises an equipment production analysis module, an equipment health assessment module, a maintenance trigger judgment module and a maintenance decision optimization terminal. According to the method, multi-modal data fusion modeling is performed and a dynamic threshold correction mechanism is constructed by integrating multi-source heterogeneous data such as vibration, temperature and voiceprint, so that the accuracy and adaptability of predictive maintenance are remarkably improved. A dynamic correlation analysis framework is constructed by combining equipment load and production environment disturbance factors, and upgrading from single feature anomaly detection to multi-modal linkage anomaly mode recognition is achieved; meanwhile, based on threshold self-correction and maintenance decision optimization of a real-time working condition, the problem of maintenance lag under a static rule is effectively solved, and a closed-loop intelligent maintenance treatment system is formed through cooperation of health index dynamic evaluation and resource scheduling.

Description

Technical Field

[0001] The present invention belongs to the field of data analysis technology, and specifically relates to a big data analysis and management system and method based on artificial intelligence. Background Art

[0002] In the context of Industry 4.0 and digital transformation, predictive maintenance in smart factories is evolving from passive response to proactive intervention through artificial intelligence and big data governance systems. Traditional maintenance models are limited by high costs, low efficiency, and data fragmentation, highlighting the need for big data dimensional analysis and governance.

[0003] Existing technologies, such as the Chinese invention patent application with application number 202310780247.9, disclose a predictive maintenance method for equipment based on operating condition timing analysis. It constructs the operating condition data topology structure through partition analysis, simplifies the monitoring dimension through cluster analysis, and realizes extremely simple prediction of faults.

[0004] Existing technologies, such as the substation equipment maintenance method based on task reliability analysis disclosed in the Chinese invention patent application with application number 202110161025.X, quantify equipment reliability fluctuations based on fault tree and sensitivity analysis, and then ensure system stability through dynamic maintenance cycle adjustment.

[0005] Obviously, both of the above solutions attempt to improve maintenance efficiency through data modeling, but the former focuses on extracting time-series correlation features, while the latter focuses on system-level reliability boundary control. In addition, there are still several shortcomings in the current predictive maintenance analysis and management of factory equipment: 1. It relies on single-modal data and lacks the ability to integrate analysis and management of multi-source heterogeneous data such as vibration and images, resulting in certain deviations in the effectiveness and reliability of predictive analysis.

[0006] 2. Currently, based on static thresholds or fixed models, it is difficult to respond to the nonlinear characteristics of equipment degradation and production environment disturbances in real time, resulting in insufficient dynamic adaptability and making it difficult to ensure the timeliness of maintenance triggering. Summary of the Invention

[0007] In view of this, in order to solve the problems raised in the above background technology, a big data analysis and governance system and method based on artificial intelligence is proposed.

[0008] The purpose of the present invention can be achieved through the following technical solutions: The present invention provides a big data analysis and governance system based on artificial intelligence, which includes the following modules: an equipment production analysis module, which dynamically adjusts the data reference range under normal operating conditions according to the usage time of the production equipment and historical fault records, and analyzes the linkage change relationship between the temperature, vibration, and voiceprint characteristics of the production equipment during the production process.

[0009] The equipment health assessment module calculates the equipment health index in the time series based on the data reference range and linkage change relationship.

[0010] The maintenance trigger judgment module corrects the maintenance trigger threshold in real time based on the current production load data and environmental data, and compares the equipment health index with the corrected maintenance trigger threshold to output the maintenance type.

[0011] The maintenance decision optimization terminal generates an optimization plan including a spare parts preparation list and maintenance time recommendations based on maintenance type, production idle period and spare parts inventory status.

[0012] The present invention also provides a big data analysis and governance method based on artificial intelligence, which includes: S1. Equipment production analysis: dynamically adjusting the data reference range under normal operating conditions according to the usage time of the production equipment and historical fault records, and analyzing the linkage change relationship between the temperature, vibration, and voiceprint characteristics of the production equipment during the production process.

[0013] S2. Equipment health assessment: Calculate the equipment health index under the time series based on the data reference range and linkage change relationship.

[0014] S3. Maintenance trigger judgment: The maintenance trigger threshold is modified in real time according to the current production load data and environmental data, and the maintenance type is output by comparing the equipment health index with the modified maintenance trigger threshold.

[0015] S4. Maintenance decision optimization: Generate an optimization plan including a spare parts preparation list and maintenance time recommendations based on maintenance type, production idle period, and spare parts inventory status.

[0016] Compared with the existing technology, the present invention has the following beneficial effects: (1) By leveraging multimodal data fusion modeling and a dynamic threshold correction mechanism, the present invention significantly improves the accuracy and adaptability of predictive maintenance. It also integrates heterogeneous data from multiple sources, such as vibration, temperature, and voiceprint, and considers equipment load and production environment disturbance factors to build a dynamic correlation analysis framework. This enables a leap from detecting only single feature anomalies to identifying multimodal linkage anomaly patterns. Furthermore, automatic threshold correction and maintenance decision optimization based on real-time working conditions effectively resolve the maintenance delay problem caused by static rules.

[0017] (2) The present invention breaks through the limitations of a single data modality by modeling the dynamic correlation of vibration, temperature, and soundprint features. It utilizes the cross-modal anomaly propagation law and intensity coupling relationship to accurately identify complex fault modes, reduce the risk of false alarms and missed alarms, provide a high-confidence basis for equipment health analysis, and ensure the effectiveness and reliability of predictive analysis results.

[0018] (3) The present invention dynamically adjusts the trigger threshold based on real-time production load and environmental parameters, adapts to the health status and external disturbance interference, avoids premature or delayed maintenance caused by traditional static thresholds, and ensures the temporal and spatial matching of maintenance instructions and the actual health status of the equipment.

[0019] (4) The present invention quantifies the equipment performance degradation rate and the cumulative effect of multi-feature anomalies by calculating the health index under time series, supports the progressive capture of early hidden faults, and provides decision-making support for the transformation of maintenance strategies from passive response to active intervention.

[0020] (5) The present invention generates a global optimal solution for maintenance timing and resource allocation by combining multiple constraints such as spare parts inventory and production idle windows, thereby reducing unplanned downtime losses and redundant spare parts occupancy, and achieving a Pareto equilibrium between maintenance costs and equipment reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 This is a schematic diagram of the system module structure connection of the present invention.

[0023] Figure 2 The figure is a schematic flow chart of the steps for implementing the method of the present invention.

[0024] Figure 3 Schematic diagram of the analysis flow of the linkage change relationship of each feature of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the 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.

[0026] See also Figure 1 As shown, the present invention provides a big data analysis and management system based on artificial intelligence, which includes: an equipment production analysis module, an equipment health assessment module, a maintenance trigger judgment module and a maintenance decision optimization terminal.

[0027] In the above, the equipment health assessment module is connected to the equipment production analysis module and the maintenance trigger judgment module respectively, and the maintenance trigger judgment module is connected to the maintenance decision optimization terminal.

[0028] The equipment production analysis module dynamically adjusts the data reference range under normal operating conditions based on the production equipment's usage time and historical fault records, and analyzes the linkage changes in the temperature, vibration, and soundprint characteristics of the production equipment during the production process.

[0029] Specifically, the adjustment of the data reference range under normal operating conditions includes: A1, predicting the remaining service life based on the usage time and historical fault records.

[0030] A2. Compare the usage time with the set safe usage time and output the planned remaining service life.

[0031] A3. If the difference between the predicted remaining useful life and the planned remaining useful life is less than 0 and exceeds the deviation range, the adjustment coefficient is assigned a value of 1. Otherwise, the ratio of the predicted remaining useful life to the safe use time is used as the adjustment coefficient.

[0032] A4. Adjust the pre-set normal value ranges of temperature, vibration, and voiceprint features by adjusting the coefficients to generate a dynamically adjusted data reference range.

[0033] The embodiment of the present invention breaks through the limitations of a single data modality by establishing a dual comparison relationship between the remaining life prediction value and the safe use threshold, and models the dynamic correlation based on vibration, temperature, and soundprint characteristics. It utilizes the cross-modal anomaly propagation law and intensity coupling relationship to accurately identify complex failure modes, reduce the risk of false alarms and missed alarms, provide a high-confidence basis for equipment health analysis, ensure the validity and reliability of the predictive analysis results, and solve the adaptability problem of monitoring parameter drift during equipment performance degradation.

[0034] Understandably, regarding step A1, the specific process of predicting the remaining service life is as follows: A11, matching each fault type with a preset fault type-impact weight table to obtain the corresponding impact weight of each fault.

[0035] It should be noted that the impact weights corresponding to fault types can be quantified based on severity, propagation speed, repair costs, and production losses. For example, bearing overheating and motor insulation failure are used as examples. For details, please refer to Table 1.

[0036] Table 1 Example of fault type-impact weight

[0037]

[0038] Regarding how to quantify the impact weight of design fault types based on factors such as severity, propagation speed, maintenance cost, and production loss, this can be achieved through the following steps and logic: extract data such as the frequency of occurrence, repair time, and loss amount of similar faults in the past, and at the same time extract the fault manual provided by the reference equipment manufacturer, the risk assessment guide issued by the industry association, or invite field experts to score typical faults, establish a quantitative standard of 0-10 points or 0-1 points, and convert the qualitative description into a numerical value. In order to be consistent with the value range of the present invention, the 0-1 point system can be used for quantification. For example, if the local functional abnormality does not affect the overall operation, the severity is recorded as mild and the severity score is set to 0.1-0.3 points. If the entire machine is paralyzed or causes a safety accident, the severity is recorded as severe and the severity score is set to 0.8-1 points. If it is between the two, the severity is recorded as moderate and the severity score is set to 0.3-0.8 points. If the abnormality does not spread for more than 24 hours, the propagation speed is recorded as slow and the propagation speed score is set to 0.1-0.3 points. If the impact reaches pre-set critical components within one hour, the propagation speed is recorded as extremely fast, and the propagation speed score is set to 0.8-1 points. If neither of these two situations occurs, the propagation speed score is set to 0.3-0.8 points. The maintenance cost score is set based on the ratio of maintenance costs to equipment purchase costs. If it exceeds the equipment purchase cost, the maintenance cost score is set to 1. If the production capacity decreases by less than 5%, the production loss is recorded as minor, and the production loss score is set to 0.1-0.3 points. If the production stoppage exceeds 48 hours, the delivery breach compensation exceeds the set warning penalty threshold, or the production capacity decrease ratio exceeds the preset warning capacity decrease ratio, the production loss is recorded as severe, and the production score is set to 0.8-1 points. Otherwise, the ratio of the capacity decrease ratio to the preset warning capacity decrease ratio is used as the production loss score. The weight coefficient of each factor is determined through expert scoring or historical data statistics, for example, severity accounts for 40%, propagation speed accounts for 20%, maintenance cost accounts for 20%, and production loss accounts for 20%. The final impact weight is finally obtained by weighted summing the scores of each factor.

[0039] The data in the table can be obtained by referring to the severity, propagation speed, repair cost and production loss recorded at the time of the corresponding fault type, and calculating them according to the above-mentioned calculation method of the final impact weight.

[0040] It should be noted that the data in the above table can be expanded according to specific industry characteristics. For example, blade cracks can be added to power generation equipment, and pipeline corrosion and other special fault types can be added to petrochemical equipment. The specific impact weight settings can also be dynamically adjusted according to industry characteristics.

[0041] A12. Calculate the average fault impact weight based on the impact weights of all faults, and calculate the relative deviation value of the impact weights between adjacent faults based on the time series of historical fault occurrences to obtain the fault impact weight change rate.

[0042] A13. Compare the timestamps of each fault occurrence to obtain the interval duration between adjacent faults, calculate the average fault interval duration, and generate the fault interval duration change rate based on the dynamic change trend analysis of the interval duration between adjacent faults.

[0043] The dynamic change trend analysis of the interval time between adjacent faults to generate the fault interval time change rate can generate a dynamic change trend curve through linear fitting, and the slope of the curve is extracted as the fault interval time change rate.

[0044] A14. Construct a fault indicator sequence based on the average fault impact weight, fault impact weight change rate, average fault interval duration, and fault interval duration change rate.

[0045] A15. Match the fault indicator sequence with a predefined fault interference factor association table, and output the matched fault interference factor.

[0046] A16. Comprehensively output the remaining service life based on the fault interference factor, usage time, and safe usage time threshold.

[0047] There is a complex mapping relationship between fault indicators and fault interference factors, which is mainly obtained by matching the fault indicator sequence with a predefined fault interference factor association table. Specifically, the fault indicator sequence constructed in step A14 contains four parameters: 、 、 and The calculation method of these parameters is described in detail in steps A12-A13. The fault interference factor association table is a pre-established mapping relationship based on the historical operation data of the equipment, industry experience, or experimental data. That is, the four parameter results simulated under various fault interference levels are pre-calculated in combination with historical operation data, etc., and the association table is constructed. That is, the fault interference factor is obtained by matching the four parameter values ​​calculated in real time with the rows in the table, and is not obtained by processing using a simple formula.

[0048] It should be added that the specific calculation formula for predicting the remaining service life is as follows: , Remaining useful life is predicted. Indicates the safe usage time threshold, Indicates the usage time. Indicates the fault interference factor. To ensure the rationality of the attenuation setting, it is necessary to set a boundary for the fault interference factor, and the fault interference factor changes dynamically with the use time.

[0049] For example, when the usage time is 10% of its safe usage time threshold, it is set to the early stage; when the usage time is less than 70% of its safe usage time threshold, it is set to the mid-term stage; when the usage time is greater than 70% of its safe usage time threshold, it is set to the aging stage. In the early stage, the equipment is new and has been strictly debugged and tested. The overall performance is stable, the various components work well together, and the tolerance to faults is good. At this time, the life prediction is not so strict. Specifically, The maximum value can be 0.1. In the medium term, as the use time increases, the parts of industrial equipment will gradually wear out and fatigue, and the risk of failure will increase. Specifically, The maximum value can be 0.2. In the aging stage, components age, accuracy decreases, faults are more likely to be triggered, and the equipment is greatly disturbed by the fault, which has a significant impact on the remaining service life. Specifically, The maximum value can be 0.5.

[0050] In another example, the average fault impact weight, the fault impact weight change rate, the average fault interval time and the fault interval time change rate are respectively recorded as 、 、 and , and assuming that it is currently in the aging stage, its specific fault indicator sequence can be seen in Table 2.

[0051] Table 2 Example of fault indicator sequence

[0052]

[0053] Understandably, what needs to be explained about step A3 is that the deviation range is a limit value determined based on the equipment's historical operating data, industry general standards, and the equipment's expected performance targets, and is used to measure the degree of deviation between the equipment's actual remaining life and the planned remaining life.

[0054] When the difference between the predicted remaining service life and the planned remaining service life is less than 0 and exceeds the deviation range, it indicates that the actual remaining service life of the equipment is far lower than expected, the equipment status has deteriorated seriously and unacceptably, and is in an emergency state that urgently needs maintenance. At this time, the adjustment coefficient is assigned a value of 1, which allows the maintenance process to quickly switch to the highest priority emergency mode to fully ensure the normal operation of the equipment.

[0055] When the difference does not show the above situation, it means that the equipment status is relatively stable. The ratio of the predicted remaining life to the safe use time is used as the adjustment coefficient. According to the ratio of the current actual remaining life of the equipment to the safe use time, the maintenance resource investment and maintenance strategy can be scientifically and reasonably adjusted to adapt to the maintenance needs of the equipment in different health conditions.

[0056] In a preferred embodiment of the present invention, the following supplement is provided regarding the adjustment performed using the adjustment coefficient in step A4: When the adjustment coefficient is assigned a value of 1, it indicates that the equipment urgently requires maintenance, and an urgent and significant adjustment is required to the normal value ranges of the integrated temperature, vibration, and voiceprint features. In this case, the originally set warning critical value range is directly used as the adjusted value range to ensure that abnormal characteristics can be effectively captured in emergency situations.

[0057] For example, assuming that the original range of the temperature difference between adjacent areas is [2°C, 5°C], the original range of the diffusion speed of the temperature anomaly area is [0.1m² / h, 0.3m² / h], and the original range of the diffusion range is [0.5m², 1m²], the corresponding adjustments are as follows: [3°C, 4°C], [0.2m² / h, 0.25m² / h], and [0.6m², 0.8m²].

[0058] The original range of the maximum amplitude difference in the vibration characteristics is [3mm, 8mm], the original range of the vibration amplitude deviation duration is [0s, 2s], and the original range of the maximum proportion is [10%, 20%]. After adjustment, they are as follows: [4mm, 6mm], [0s, 1s], and [12%, 16%] respectively.

[0059] The original range of the maximum decibel value in the voiceprint feature is [60dB, 80dB], and the original range of the abnormal duration is [0s, 3s]. After adjustment, they are as follows: [65dB, 75dB] and [0s, 1.5s].

[0060] When the adjustment coefficient is the ratio of the predicted remaining life to the safe use time, the equipment status is relatively stable, and the normal value range of each feature is dynamically adjusted based on this.

[0061] For example, the specific adjustment formula for the lower limit of the normal characteristic value range is: , Indicates the lower limit of the adjusted value range. and Respectively represent the upper and lower limits of the normal value range, represents the adjustment coefficient, Indicates the adjustment coefficient of the lower limit value, which can be set to 0.4.

[0062] The specific adjustment formula for the upper limit of the normal value range of the characteristic is: , Indicates the upper limit of the adjusted value range. Indicates the adjustment coefficient of the upper limit value, which can be set to 0.3.

[0063] Taking temperature difference as an example, when the adjustment coefficient is 0.8, the original normal range of temperature difference between adjacent areas is [1.5°C, 4.5°C]. The lower limit of the temperature difference between adjacent areas is adjusted to 1.5 + (4.5 − 1.5) × (1 − 0.8) × 0.4 = 1.74°C, and the upper limit is adjusted to 4.5 − (4.5 − 1.5) × (1 − 0.8) × 0.4 = 4.26°C.

[0064] For more details, please refer to Figure 3 As shown, the specific analysis process of the linkage change relationship is as follows: B1. Through infrared thermal imagers, vibration sensors and voiceprint collection devices, infrared images, vibration signals and voiceprint signals of each collection position of production equipment are obtained in real time, and feature extraction and standardization processing are performed.

[0065] B2. Within the same time window, check whether the temperature, vibration, and voiceprint features at the same location exceed the preset threshold. If at least two features exceed the threshold at the same time, mark it as a synchronous abnormal event and record the abnormal feature combination.

[0066] B3. Count the abnormal intensity and direction of the three features in the synchronous abnormal events to build an abnormal feature association matrix, and map the synchronous abnormal events to the corresponding physical locations of the equipment to generate an abnormal heat map.

[0067] B4. If the anomalies of the three characteristics have a chronological order, use time series analysis to determine the anomaly propagation path, and construct a multi-level causal chain based on the temporal relationship. Then, mark the starting point and propagation path of the causal chain in the anomaly heat map.

[0068] It should be added that when the anomalies of the three features do not have a sequential order and at least two features do not exceed the threshold at the same time, it indicates that there is no linkage relationship, and you can return to step B1 to collect data again.

[0069] It can be understood that the feature extraction in step B1 includes: B11, identifying areas in the infrared image where the temperature exceeds a set threshold, recording them as temperature abnormality areas, and marking their positions.

[0070] For example, the temperature threshold can be determined based on historical temperature data during normal operation of the equipment, industry standards, and equipment design parameters. For example, for a specific piece of equipment, the normal operating temperature range is 30°C-50°C. After long-term monitoring and data analysis, areas with temperatures above 60°C are defined as temperature anomaly areas.

[0071] B12. Compare and analyze the diffusion speed and range of the temperature abnormality area by continuous multi-frame image analysis.

[0072] Among them, a specific example of analyzing the diffusion speed of the temperature anomaly area is: select multiple consecutive frames of images, assuming they are the nth frame, the n+1th frame, the n+2th frame... For two adjacent frames of images, such as the nth frame and the n+1th frame, first determine the centroid position of the temperature anomaly area in the two frames. The centroid position can be obtained by calculating the weighted average of the coordinates of each pixel in the temperature anomaly area, and the weight is the temperature value of the pixel. The higher the temperature, the greater the weight. Suppose the centroid coordinates of the temperature anomaly area in the nth frame are ( , ), the centroid coordinates of the temperature anomaly area in the n+1th frame are ( , ), the time interval between two frames of image acquisition is ,exist Axial velocity component The calculation formula is: ,exist Axial velocity component The calculation formula is: , and then the actual moving speed of the center of mass is calculated using the Pythagorean theorem. This speed value can approximately represent the diffusion speed of the temperature anomaly area during this time period. To obtain a more accurate diffusion speed, the diffusion speeds between multiple groups of adjacent frames can be averaged.

[0073] Furthermore, when analyzing the diffusion range of the temperature anomaly area, the pixel counting method is first used to count the number of pixels in the temperature anomaly area in each frame of the image. The number of pixels is converted into the actual area based on the conversion relationship between the pixels of the thermal imaging device and the actual area. It is assumed that the actual area represented by each pixel is , the number of pixels in the temperature anomaly area is , then the actual area of ​​the region is , , for example, the area of ​​the temperature anomaly region in the nth frame is , the area of ​​the temperature anomaly region in the n+j frame is ,when When , its diffusion range is .

[0074] B13. Calculate the temperature difference between adjacent areas and integrate the temperature difference between adjacent areas, the diffusion speed and range of the temperature anomaly area into temperature characteristics.

[0075] B14. Perform a difference analysis based on the maximum amplitude value of the vibration waveform in the vibration signal and the preset fluctuation amplitude value to obtain the maximum amplitude value difference. At the same time, calculate the duration of the amplitude value difference being greater than the fluctuation amplitude value difference, which is recorded as the vibration amplitude deviation duration.

[0076] B15. Use frequency domain analysis to calculate the proportion of high-frequency and low-frequency components in the vibration signal, extract the maximum proportion, and use the maximum amplitude difference, duration of vibration amplitude deviation, and maximum proportion as vibration characteristics.

[0077] The energy of each frequency band can be approximately calculated by summing the squares of the amplitudes of each frequency point in the band. The proportion of high-frequency components is the ratio of the energy of the high-frequency band to the total energy of each frequency band, and the proportion of low-frequency components is the ratio of the energy of the low-frequency band to the total energy of each frequency band.

[0078] B16. Extract the maximum decibel value of the voiceprint signal, and count the duration of the decibel value exceeding the normal noise, which is recorded as the abnormal duration. Then, the maximum decibel value and the abnormal duration are used as voiceprint features.

[0079] Understandably, the specific determination of the anomaly propagation path using time series analysis in step B4 is as follows: Using time series analysis methods such as the Autoregressive Integrated Moving Average (ARIMA) model, the three characteristic anomaly time series are modeled separately to analyze their changing trends and patterns. For example, the ARIMA model can be used to analyze the temporal trend of temperature anomalies to determine whether the temperature anomaly is gradually increasing or suddenly surging. The ARIMA model is an existing model, so the detailed analysis process is not detailed here.

[0080] In a preferred embodiment of the present invention, in step B4, the construction of a multi-level causal relationship chain and the marking of the starting point and propagation path of the causal chain are specifically demonstrated using an example of a failure caused by equipment overheating. The specific example is as follows: Abnormal temperature characteristics at all times, local overheating, The vibration amplitude suddenly increases at this moment. High-frequency noise appears in the voiceprint spectrum at this moment. , , as well as Respectively Moment and The time interval and Moment and The duration of the time interval.

[0081] The first-level causal relationship is temperature anomaly → vibration anomaly, with a Granger causality test P value of 0.01 and a transfer entropy value of 0.35. The second-level causal relationship is vibration anomaly → voiceprint anomaly, with a transfer entropy value of 0.28. The resulting multi-level causal chain is temperature anomaly → vibration anomaly → voiceprint anomaly.

[0082] Take temperature anomaly as the starting point of the causal chain and mark it as the risk path of "overheating-vibration-noise".

[0083] The Granger causality test is a common statistical method used to determine whether variable A Granger causes variable B. Specifically, it determines whether variable A's historical information significantly predicts the current value of variable B. Regarding the calculation logic for the P value of 0.01, transfer entropy of 0.35, and transfer entropy of 0.28 involved in verifying the causal relationship between "temperature anomaly → vibration anomaly" and "vibration anomaly → voiceprint anomaly," please refer to the technical principles mentioned in this article and the following supplementary explanation of existing technologies:

[0084] P-value calculation: The temperature values ​​exceeding the preset temperature threshold are combined into a temperature anomaly value sequence, and the vibration values ​​exceeding the preset threshold are combined into a vibration anomaly value sequence. Stationarity tests such as the ADF test are performed on the temperature anomaly value sequence and the vibration anomaly value sequence. If they are not stationary, they are converted into stationary series through differencing. A univariate ARIMA model is constructed for the vibration anomaly value sequence, in which the sequence depends only on its own historical values, and a multivariate ARIMA model is constructed for the vibration anomaly value sequence, in which the sequence depends on both its own historical values ​​and the historical values ​​of the temperature anomaly value sequence.

[0085] Compare the sum of squares of the prediction errors of the univariate ARIMA model with the multivariate ARIMA model and calculate the F statistic. ,in, represents the sum of squared forecast errors of the univariate ARIMA model, represents the sum of squared forecast errors of the multivariate ARIMA model, is the independent variable, such as the lag order of temperature, is the dependent variable such as the lag order of vibration, is the sample size, and the P value refers to the probability of observing the current F value and more extreme values ​​under the premise that the original hypothesis, that is, there is no causal relationship between temperature anomaly and vibration anomaly, is established. The F value obeys the degree of freedom of The F distribution can be used to calculate the right-tail probability corresponding to the F value, that is, the P value.

[0086] Transfer entropy calculation: The transfer entropy calculation formula is: , which is used to measure the historical state of the known variable Y When the current state of variable X The future state of Y The predicted contribution of .

[0087] In combination with the content of the present invention, the calculation of transfer entropy needs to be based on the joint probability distribution of time series, and the continuous temperature, vibration, and voiceprint feature values ​​are divided into discrete states such as low, medium, and high to facilitate the calculation of probability distribution. is the temperature or vibration state at time t-1, is the temperature or vibration state at time t, The vibration or voiceprint state at time t is calculated by historical data and the following probabilities are obtained: 、 and , substitute into the transfer entropy formula to calculate the transfer entropy value.

[0088] It should be noted that the Granger causality test and transfer entropy confirmed the causal relationship between abnormal temperature and abnormal vibration. The P value of the Granger causality test was 0.01, which is lower than the common significance level of 0.05. This indicates that abnormal temperature has a statistically significant impact on abnormal vibration, indicating that abnormal temperature is a cause of abnormal vibration. The transfer entropy was 0.35, indicating the degree of information transfer between the two variables. In other words, the abnormal temperature transmits a certain amount of information to the abnormal vibration, further supporting this causal relationship.

[0089] Although Granger causality tests were not performed for the second-level causal relationship, the causal relationship between the vibration anomaly and the voiceprint anomaly was confirmed using transfer entropy. A transfer entropy of 0.28 indicates that the vibration anomaly transmits some information to the voiceprint anomaly, meaning that the vibration anomaly is a cause of the voiceprint anomaly. A transfer entropy value of 0 indicates that there is no information transfer between the two variables and that they are independent of each other. A transfer entropy value greater than 0 indicates that the state of one variable can provide predictive information for the other, indicating that information transfer occurs.

[0090] The equipment health assessment module calculates the equipment health index in the time series according to the data reference range and the linkage change relationship.

[0091] Specifically, the device health index analysis includes: D1, calculating the deviation of temperature, vibration and voiceprint features within each time window relative to the data reference range.

[0092] It can be understood that the deviation is obtained by calculating the difference between the temperature, vibration and voiceprint characteristic values ​​in each time window and the corresponding data reference range, and comparing it with the amplitude of the reference range.

[0093] D2. Setting the device linkage deviation based on the linkage change relationship, and analyzing the comprehensive device deviation according to the deviation of each feature and the device linkage deviation.

[0094] D3. If the comprehensive deviation is 0, the upper limit of the preset device health index reference range is used as the device health index. Otherwise, the inverse of the comprehensive deviation is mapped to the preset device health index reference range through the Sigmoid function, and the mapped value is used as the device health index.

[0095] It's important to note that a comprehensive deviation of 0 means that all device characteristics are completely consistent with normal standards, with no deviations. At this point, the device is operating in ideal, normal conditions. In this case, the upper limit of the preset device health index reference range can be used as the device health index. Conversely, a lower value indicates a less ideal device operating condition and a higher potential failure risk. The Sigmoid function compresses input values ​​into the (0, 1) range, allowing it to effectively compress and convert the inverse of comprehensive deviations over a wide range.

[0096] D4. Bind the device health index of each time window to the window end timestamp and output the device health index in the time series.

[0097] The embodiment of the present invention quantifies the equipment performance degradation rate and the cumulative effect of multi-feature anomalies by calculating the health index under time series, supports the progressive capture of early hidden faults, and provides decision-making support for the transformation of maintenance strategies from passive response to active intervention.

[0098] Furthermore, the process of setting the device linkage deviation in step D2 includes: D21, calculating the linkage strength between temperature, vibration and voiceprint features by using the Pearson correlation coefficient to generate a linkage weight.

[0099] It should be noted that the Pearson correlation coefficient is an existing function and its specific calculation formula is no longer shown.

[0100] D22. If a synchronous abnormal event is triggered, the abnormal trigger weight is obtained by matching the preset weight table based on the abnormal feature combination, and the final weight is analyzed based on the abnormal trigger weight and linkage weight.

[0101] It should be noted that the preset weight table is pre-defined based on a large amount of historical data and domain expert experience. This table assigns corresponding weights to different abnormal feature combinations, reflecting the degree of impact of each abnormal feature combination on equipment failure or abnormal status.

[0102] D23. If no abnormal event is triggered, perform Granger causality test on the multi-level causal chain, assign causal weights according to the absolute value ratio of the Granger coefficient, and generate the final weight by combining the linkage weight and causal weight.

[0103] It should be noted that the Granger causality test is a common technical means currently available, and its specific test process will not be repeated here. The final weight generated by combining the linkage weight and causal weight can be obtained by calculating the average value.

[0104] D24. Calculate the output device linkage deviation by weighted summation based on the final weight and the deviation of each feature.

[0105] The maintenance trigger judgment module amends the maintenance trigger threshold in real time according to the current production load data and environmental data, and compares the equipment health index with the amended maintenance trigger threshold to output the maintenance type.

[0106] Specifically, the process of real-time correction of the maintenance trigger threshold comprises the following steps: H1, extracting equipment utilization and production plan from current production load data, and comprehensively analyzing and calculating production urgency.

[0107] H2. Calculate the corresponding deviation between the current environmental data and the adapted environmental data under normal operating conditions of the equipment to obtain the environmental deviation degree.

[0108] It should be added that the calculation principle of the environmental deviation is the same as the calculation principle of the deviation in the above step D1, and no further explanation is given.

[0109] H3. Generate a correction factor after normalizing the production urgency and environmental deviation, and adjust the preset maintenance trigger threshold based on the correction factor.

[0110] Furthermore, the specific output process of the maintenance type includes: E1, matching the device health index in the current time window with the triggering device health index threshold corresponding to each maintenance type, and if the match is successful, outputting the matching maintenance type.

[0111] E2. Otherwise, construct a device health index change curve with the time window as the horizontal axis and the device health index as the vertical axis, and calculate the slope of the curve as the device health index change rate.

[0112] E3. If the equipment health index change rate is less than 0 and its absolute value exceeds the preset warning value, the output component detection maintenance type.

[0113] The embodiments of the present invention dynamically adjust the trigger threshold based on real-time production load and environmental parameters, adapt to the health status and external disturbance interference, avoid premature or delayed maintenance caused by traditional static thresholds, and ensure the temporal and spatial matching of maintenance instructions and the actual health status of the equipment.

[0114] The maintenance decision optimization terminal generates an optimization plan including a spare parts preparation list and maintenance time recommendations based on the maintenance type, production idle period and spare parts inventory status.

[0115] The embodiment of the present invention generates a global optimal solution for maintenance timing and resource allocation by combining multiple constraints such as spare parts inventory and production idle windows, reducing unplanned downtime losses and redundant spare parts occupancy, and achieving a Pareto equilibrium between maintenance costs and equipment reliability.

[0116] Exemplarily, the optimization solution is specifically generated as follows: reviewing the production schedule, screening suitable continuous idle windows, and checking the spare parts inventory status to determine the inventory quantity, availability, and replenishment cycle of the required spare parts.

[0117] If spare parts are available and a production downtime period meets the required maintenance duration, a spare parts preparation list is generated based on the maintenance type. Preventive maintenance lists for routine wear parts also use production downtime as maintenance time. Component inspection and maintenance lists include specialized inspection tools and small test replacement parts, with maintenance times planned based on inspection processes and duration.

[0118] If spare parts need to be restocked, a plan should be developed based on the replenishment cycle and subsequent production downtime. Preventive maintenance should be based on the regular replenishment cycle for consumable parts, coordinated with subsequent downtime, and clearly defined procurement time and transportation methods. For component inspection and maintenance, consider the difficulty and cycle of specialized spare parts replenishment. Communicate with suppliers in advance about customized or expedited procurement, adjust maintenance schedules based on the new timeframe, and plan inspection and component replacement procedures.

[0119] The embodiments of this invention significantly improve the accuracy and adaptability of predictive maintenance by leveraging multimodal data fusion modeling and a dynamic threshold correction mechanism. They also integrate heterogeneous data from multiple sources, such as vibration, temperature, and voiceprints, and build a dynamic correlation analysis framework that considers equipment load and production environment disturbances. This advances the process from detecting only single-feature anomalies to identifying multimodal, linked anomaly patterns. Furthermore, automatic threshold correction and maintenance decision optimization based on real-time operating conditions effectively address the maintenance delays caused by static rules.

[0120] See also Figure 2 As shown, the present invention also provides a big data analysis and governance method based on artificial intelligence, which includes: S1, equipment production analysis: dynamically adjusting the data reference range under normal operating conditions according to the usage time of the production equipment and historical fault records, and analyzing the linkage change relationship of the temperature, vibration, and voiceprint characteristics of the production equipment during the production process.

[0121] S2. Equipment health assessment: Calculate the equipment health index under the time series based on the data reference range and linkage change relationship.

[0122] S3. Maintenance trigger judgment: The maintenance trigger threshold is modified in real time according to the current production load data and environmental data, and the maintenance type is output by comparing the equipment health index with the modified maintenance trigger threshold.

[0123] S4. Maintenance decision optimization: Generate an optimization plan including a spare parts preparation list and maintenance time recommendations based on maintenance type, production idle period, and spare parts inventory status.

[0124] The above formula is a formula obtained by collecting a large amount of data and performing software simulation to obtain the latest real situation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0125] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0126] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0127] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0128] 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.

[0129] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. Big data analysis and management system based on artificial intelligence, characterized by: The system includes the following modules: The equipment production analysis module dynamically adjusts the data reference range under normal operating conditions based on the equipment's usage time and historical fault records, and analyzes the linkage changes in the temperature, vibration, and soundprint characteristics of the equipment during the production process; The equipment health assessment module calculates the equipment health index in time series based on the data reference range and linkage change relationship; The maintenance trigger judgment module adjusts the maintenance trigger threshold in real time based on the current production load data and environmental data, and compares the equipment health index with the adjusted maintenance trigger threshold to output the maintenance type; The maintenance decision optimization terminal generates an optimization plan including a spare parts preparation list and maintenance time recommendations based on maintenance type, production idle period and spare parts inventory status.

2. The big data analysis and management system based on artificial intelligence according to claim 1 is characterized by: The reference range of data under normal operating conditions of the adjustment includes: Predict remaining useful life based on usage time and historical failure records; Compare the usage time with the set safe usage time and output the planned remaining service life; If the difference between the predicted remaining service life and the planned remaining service life is less than 0 and exceeds the deviation range, the adjustment coefficient is assigned a value of 1; otherwise, the ratio of the predicted remaining service life to the safe use time is used as the adjustment coefficient; The preset normal value ranges of temperature, vibration and voiceprint features are adjusted by adjusting the coefficients to generate a dynamically adjusted data reference range.

3. The big data analysis and management system based on artificial intelligence according to claim 2 is characterized by: The specific process of predicting the remaining service life is as follows: Match each fault type with the preset fault type-impact weight table to obtain the corresponding impact weight of each fault; The average fault impact weight is calculated based on the impact weights of all faults, and the relative deviation value between the impact weights of adjacent faults in the time series of historical fault occurrence is calculated to obtain the fault impact weight change rate; Compare the timestamps of each fault occurrence to obtain the interval between adjacent faults, calculate the average interval between faults, and generate the rate of change of the interval between faults based on the dynamic trend analysis of the interval between adjacent faults. Construct a fault indicator sequence based on the average fault impact weight, fault impact weight change rate, average fault interval time and fault interval time change rate; Match the fault indicator sequence with the predefined fault interference factor association table and output the matched fault interference factor; The remaining service life is predicted by comprehensively outputting the fault interference factor, usage time and safe usage time threshold.

4. The big data analysis and management system based on artificial intelligence according to claim 1 is characterized by: The specific analysis process of the linkage change relationship is as follows: Through infrared thermal imagers, vibration sensors and voiceprint collection devices, infrared images, vibration signals and voiceprint signals at each collection position of production equipment are acquired in real time, and feature extraction and standardization are performed; Within the same time window, check whether the temperature, vibration, and voiceprint features at the same location exceed the preset threshold. If at least two features exceed the threshold at the same time, it is marked as a synchronous abnormal event and the abnormal feature combination is recorded; The abnormal intensity and direction of the three features in the synchronous abnormal events are counted to construct an abnormal feature correlation matrix, and the synchronous abnormal events are mapped to the corresponding physical locations of the equipment to generate an abnormal heat map; If the anomalies of the three characteristics have a sequence, time series analysis is used to determine the anomaly propagation path, and a multi-level causal chain is constructed based on the temporal relationship, and then the starting point and propagation path of the causal chain are marked in the anomaly heat map.

5. The big data analysis and management system based on artificial intelligence according to claim 4 is characterized by: The feature extraction includes: Identify areas in infrared images where the temperature exceeds a set threshold, record them as abnormal temperature areas, and mark their locations; Compare and analyze the diffusion speed and range of the temperature abnormality area by continuous multi-frame image analysis; Calculate the temperature difference between adjacent areas, and integrate the temperature difference between adjacent areas, the diffusion speed and range of the temperature anomaly area into temperature characteristics; The maximum amplitude difference is obtained by performing a difference analysis based on the maximum amplitude value of the vibration waveform in the vibration signal and the preset fluctuation amplitude value. At the same time, the duration of the amplitude difference being greater than the fluctuation amplitude difference is counted and recorded as the duration of the vibration amplitude deviation; Through frequency domain analysis, the proportion of high-frequency and low-frequency components in the vibration signal is statistically analyzed, and the maximum proportion is extracted. The maximum amplitude value difference, the duration of vibration amplitude deviation, and the maximum proportion are used as vibration features; The maximum decibel value of the voiceprint signal is extracted, and the duration of the decibel value exceeding the normal noise is counted, which is recorded as the abnormal duration. The maximum decibel value and the abnormal duration are then used as voiceprint features.

6. The big data analysis and management system based on artificial intelligence according to claim 4 is characterized by: The equipment health index analysis includes: Calculate the deviation of temperature, vibration and voiceprint features within each time window relative to the data reference range; Setting the device linkage deviation based on the linkage change relationship, and analyzing the comprehensive device deviation according to the deviation of each feature and the device linkage deviation; If the comprehensive deviation is 0, the upper limit of the preset device health index reference range is used as the device health index. Otherwise, the inverse of the comprehensive deviation is mapped to the preset device health index reference range through the Sigmoid function, and the mapped value is used as the device health index. Bind the device health index of each time window to the window end timestamp and output the device health index in the time series.

7. The big data analysis and management system based on artificial intelligence according to claim 6 is characterized by: The process of setting the device linkage deviation includes: The linkage strength between temperature, vibration and voiceprint features is calculated through the Pearson correlation coefficient to generate linkage weights; If a synchronous abnormal event is triggered, the abnormal trigger weight is obtained by matching the preset weight table based on the abnormal feature combination, and the final weight is analyzed based on the abnormal trigger weight and linkage weight; If no abnormal event is triggered, Granger causality test is performed on the multi-level causal relationship chain, causal weights are assigned according to the absolute value ratio of the Granger coefficient, and the final weight is generated by combining the linkage weight and causal weight; The output device linkage deviation is calculated based on the final weight and the deviation of each feature.

8. The big data analysis and management system based on artificial intelligence according to claim 1 is characterized by: The process of real-time correction and maintenance trigger threshold includes: Extract equipment utilization and production plan from current production load data, and perform comprehensive analysis to calculate production urgency; The current environmental data and the adapted environmental data under normal operating conditions of the equipment are compared to calculate the corresponding deviation to obtain the environmental deviation degree; A correction factor is generated according to production urgency and environmental deviation, and the preset maintenance trigger threshold is adjusted based on the correction factor.

9. The big data analysis and management system based on artificial intelligence according to claim 1 is characterized by: The specific output process of the maintenance type includes: Match the device health index in the current time window with the trigger device health index threshold corresponding to each maintenance type. If a match is successful, output the matching maintenance type; Otherwise, construct a device health index change curve with the time window as the horizontal axis and the device health index as the vertical axis, and calculate the slope of the curve as the device health index change rate; If the equipment health index change rate is less than 0 and its absolute value exceeds the preset warning value, the output component detects the maintenance type.

10. An artificial intelligence-based big data analysis and governance method, characterized by: The method includes: S1. Equipment production analysis: Dynamically adjust the data reference range under normal operating conditions based on the production equipment's usage time and historical fault records, and analyze the linkage changes in the temperature, vibration, and soundprint characteristics of the production equipment during the production process; S2. Equipment health assessment: Calculate the equipment health index in time series based on the data reference range and linkage change relationship; S3. Maintenance trigger judgment: The maintenance trigger threshold is modified in real time based on the current production load data and environmental data, and the maintenance type is output by comparing the equipment health index with the modified maintenance trigger threshold; S4. Maintenance decision optimization: Generate an optimization plan including a spare parts preparation list and maintenance time recommendations based on maintenance type, production idle period, and spare parts inventory status.

Images