Centrifugal fan fault trend prediction method based on multi-source information fusion

Abstract

The invention belongs to the technical field of centrifugal fan fault prediction, and provides a centrifugal fan fault trend prediction method based on multi-source information fusion, and the method comprises the following steps: in the operation process of a centrifugal fan, capturing the change trend of data distribution through multi-source data monitoring and statistical property analysis; and whether data non-stationary change of the centrifugal machine is caused by equipment aging is identified. Based on a time and space two-dimensional comparison analysis framework, physical mechanism verification is combined, suspicious aging features are locked by calculating relative deviation and aging trend index quantitative indexes, and then an average correlation coefficient and trend consistency are utilized to construct a correlation consistency index to verify feature reliability, so that the method has the capability of positioning a specific aging part, and the reliability of the aging part is improved. The problem of misjudgment caused by sensor faults or environmental interference in traditional fault diagnosis is optimized, the accuracy of aged part recognition is improved, and a clear object is provided for targeted maintenance.

Description

Technical Field

[0001] The present invention belongs to the technical field of centrifugal fan fault prediction, and in particular relates to a centrifugal fan fault trend prediction method based on multi-source information fusion. Background Art

[0002] Centrifugal fans, as key equipment in industrial production, are widely used in ventilation, dust removal, pneumatic conveying, and other scenarios. Their long-term stable operation is crucial to production continuity and safety. During long-term operation, core components of the equipment (such as bearings, impellers, and motors) inevitably age due to factors such as friction, corrosion, and fatigue. This can manifest as bearing wear, impeller corrosion, and winding aging. These aging processes are not sudden but rather accompany slow changes in the statistical characteristics of the data. For example, bearing wear can cause the vibration mean to gradually increase, while impeller corrosion can cause the wind pressure baseline to gradually decrease. This gradual change in statistical characteristics over time is called non-stationarity and is the core characteristic of equipment degradation. Traditional centrifugal fan fault trend prediction methods are mostly based on the assumption of stationary data. This assumes that the statistical characteristics of equipment operating data (such as mean and variance) remain stable, and model training relies on historical health data. However, in actual operation, the non-stationarity caused by equipment aging can lead to continuous changes in data distribution. For example, the mean value of bearing vibration increases from its healthy value. This slow and continuous deviation cannot be effectively captured by traditional static models. When the non-stationarity of multi-source data is not identified and addressed, the model will continue to rely on outdated health data features for prediction and will be unable to track the dynamic changes caused by aging. This causes the model's understanding to lag behind the actual development of the fault, resulting in the prediction of the fault occurrence time and severity lagging behind the actual equipment status, missing the optimal maintenance opportunity. To this end, the present invention provides a centrifugal fan fault trend prediction method based on multi-source information fusion. Summary of the Invention

[0003] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0004] The technical solution adopted by the present invention to solve the technical problem is: a method for predicting the fault trend of a centrifugal fan based on multi-source information fusion, comprising the following steps: During the operation of the centrifugal fan, by monitoring multi-source data and analyzing statistical characteristics, the changing trend of data distribution is captured, and it is identified whether the centrifuge has caused non-stationary data changes due to equipment aging; If the non-stationary changes are caused by equipment aging, the initial health status data of the equipment is used as the baseline value, and a comparative analysis of the two dimensions of time and space is performed; The process of the dual-dimensional comparison analysis of time and space is as follows: first, for the features corresponding to non-stationary changes, the relative deviation of each data point is calculated within a preset time window. By analyzing the relative deviation, an aging trend index is outputted. The aging trend index is used to extract suspicious aging features. Then, for the suspicious aging components mapped by the suspicious aging features, the correlation consistency index of the sensor in the healthy state is calculated. The correlation consistency index is used to identify the reliability of the suspicious aging components corresponding to the suspicious aging features. If reliable, then combine the physical mechanism to verify the target component whose output has aged; For target components that are aging, a dynamic updated reinforcement learning model strategy is automatically switched to predict fault trends. The reinforcement learning model includes setting a dynamic sliding data window and introducing a healthy forgetting factor.

[0005] The beneficial effects of the present invention are as follows: This method monitors and processes multi-source data during centrifugal fan operation in real time. It uses the ADF unit root test to identify non-stationary changes in data caused by equipment aging, thereby capturing the characteristics of progressive equipment aging. This method overcomes the difficulty of traditional methods in distinguishing between true aging and random interference, laying a precise foundation for early judgment of subsequent fault trend predictions. Based on a dual-dimensional comparison and analysis framework of time and space, combined with physical mechanism verification, this invention locates suspicious aging features by calculating relative deviation and aging trend index quantitative indicators. It then uses the average correlation coefficient and trend consistency to construct an associated consistency index to verify the reliability of the features. This allows it to locate specific aging components, optimizes the misjudgment problem caused by sensor failure or environmental interference in traditional fault diagnosis, improves the accuracy of aging component identification, and provides clear targets for targeted maintenance. The present invention is based on a dynamically updated reinforcement learning model strategy, introduces a dynamic sliding data window and a health forgetting factor, enables the model to incorporate new data in real time and weaken the influence of outdated health data, thereby having the ability to adapt to the aging evolution trend of adaptive components, optimizes the prediction lag problem of static models for non-stationary data, improves the prediction accuracy of fault occurrence time and severity, and provides more scientific decision support for the preventive maintenance of centrifugal fans. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The present invention will be further described below with reference to the accompanying drawings.

[0007] Figure 1 This is a flowchart of the steps of a centrifugal fan fault trend prediction method based on multi-source information fusion according to the present invention; Figure 2 This is a flow chart of step S20 in a centrifugal fan fault trend prediction method based on multi-source information fusion according to the present invention. DETAILED DESCRIPTION

[0008] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0009] Example 1 See also Figure 1 and Figure 2 As shown, a centrifugal fan fault trend prediction method based on multi-source information fusion according to an embodiment of the present invention includes the following steps: Step S10: During the operation of the centrifugal fan, through multi-source data monitoring and statistical characteristic analysis, the changing trend of data distribution is captured to identify whether the centrifuge has caused non-stationary data changes due to equipment aging; During long-term operation of a centrifuge, equipment aging (such as bearing wear and impeller corrosion) can cause slow changes in data statistical characteristics (such as a gradual increase in the vibration mean). The essence of this phenomenon is the "non-stationarity" of the equipment state, that is, the system dynamic characteristics gradually change over time. The core purpose of this step is to identify this non-stationarity phenomenon. In some embodiments, during the operation of the centrifugal fan, sensors are provided on core components of the centrifugal fan to collect multi-source data in real time, wherein the core components include but are not limited to: bearings, impellers, motors, and air inlets; Among them, multi-source data includes but is not limited to: vibration data, temperature data, pressure and flow data, and operating condition data; All data is transmitted to the data acquisition terminal via an industrial bus (such as Modbus and Profinet), and stored after being aligned by timestamp, forming a time series database containing time-multi-source features; Preprocess the acquired multi-source data, including data cleaning and standardization, and eliminating noise and interference; For the multi-source data after preprocessing, a time window is set and the features of statistical characteristics are extracted within the time window to capture the trend changes of multi-source data over time; Among them, statistical characteristics include: basic statistics, frequency domain characteristics and trend indicators; Specifically, basic statistics include but are not limited to: calculating the mean, variance, and maximum value of the vibration signal within each time window, the mean and heating rate of the temperature signal, and the baseline value and fluctuation amplitude of the pressure signal; Frequency domain features include but are not limited to: Fourier transform of vibration signals to extract the power spectrum density of bearing characteristic frequencies and the energy proportion of impeller blade pass frequencies, reflecting the wear or imbalance of the mechanical structure; Trend indicators include: calculating the slope of features in a continuous window, for example, the daily growth rate of the vibration mean, the weekly decay rate of the pressure baseline, and the rate of change of quantitative statistical characteristics; The ADF unit root test method is used to analyze the multi-source data time series to identify whether the statistical characteristics have changed significantly over time. The process is as follows: The ADF unit root test method is used to determine whether a time series has a "unit root": if the sequence has a unit root, it is a non-stationary sequence, and its statistical characteristics change over time, which is consistent with the non-stationary characteristics caused by equipment aging; if there is no unit root, it is a stationary sequence, with stable statistical characteristics and no significant aging-driven trend changes; A regression model with ADF test is established for the statistical characteristic sequence of multi-source data of centrifugal fans. For example, the multi-source data statistical feature series of a centrifugal fan are: vibration mean-time series, bearing characteristic frequency PSD-time series; The form of the regression model is: ;in, is the statistical characteristic value at time t; is the first-order difference of the sequence; is a constant term; is the time trend term; is the trend term coefficient, is the coefficient to be estimated, is the eigenvalue lagged one period, is the lagged difference term; is the random error term, k is the lag order; Set the test hypothesis: Null hypothesis: the sequence has a unit root (γ=0), that is, the sequence is non-stationary; Alternative hypothesis: There is no unit root in the sequence (γ < 0), that is, the sequence is stationary; The parameters of the regression model are estimated by the least squares method, and the ADF statistic, which is the test indicator of the estimated parameters, is calculated; Compare the ADF statistic with a predetermined significance level, which is determined by the theoretical distribution of the ADF test. If the ADF statistic is less than the critical value, the null hypothesis is rejected and the series is determined to be stationary; If the ADF statistic is greater than or equal to the critical value, the null hypothesis cannot be rejected and the series is determined to be non-stationary; It should be noted that the above ADF test method can only determine whether the series is stationary. It is necessary to combine the significance of the trend term in the regression model to determine whether the data is non-stationary due to equipment aging. The process is as follows: If the sequence is judged to be non-stationary and the trend term is significant, that is, the t-test is passed and the P value is less than 0.05; The trend term is greater than zero, and combined with the physical mechanism verification, it is determined to be a non-stationary change caused by the equipment; For example, the physical mechanism is that if the trend term of the vibration mean series is positive, it conforms to the physical law that bearing wear will gradually increase the vibration mean; The trend term is less than zero, and combined with the physical mechanism verification, it is determined to be a non-stationary change caused by the equipment; For example, the physical mechanism is that if the trend term of the pressure baseline sequence is negative, it is consistent with the characteristic that the wind pressure gradually decreases due to impeller corrosion; If the sequence is determined to be non-stationary and the trend term is not significant, that is, the t-test passes with a P value ≥ 0.05, then it is not determined to be non-stationary due to equipment-induced changes. Non-stationarity may be caused by random fluctuations, sensor noise, or sudden interference other than aging. For example, the non-stationarity determination of the vibration mean sequence caused by bearing wear is performed; After three years of operation, the centrifugal fan bearing #1 (drive end) was subjected to a vibration sensor to collect X-direction vibration acceleration signals. After preprocessing, the signals were converted into a "vibration mean-time" series (in g). The sampling interval was one hour, and the data totaled 30 days. ADF unit root test: Construct a regression model for the "vibration mean-time" series and determine the lag order k=2 using the AIC criterion; The calculated ADF statistic is -2.35, and the critical value at the 5% significance level is -3.43; Since the ADF statistic (-2.35) is greater than or equal to the critical value (-3.43), the null hypothesis cannot be rejected and the sequence is determined to be non-stationary. The t-test of the trend term coefficient was performed, and the calculated P value was 0.02 (<0.05), indicating that the trend term was significant; The estimated trend coefficient is 0.015 (>0), which means that the mean vibration value has a significant upward trend over time (an average daily increase of 0.015g). The physical mechanism is that bearing wear will cause the gap between the rolling elements and the inner and outer rings to increase, and the friction impact will intensify. This is directly manifested as the vibration mean gradually increasing over time, which is consistent with the statistical result of "trend item > 0"; It is determined that the non-stationarity of the vibration mean-time series is caused by the aging (wear) of bearing #1; Step S20: If the non-stationary change is caused by equipment aging, then a time and space dual-dimensional comparison analysis is performed based on the equipment's initial health status data, and the target component experiencing aging is verified and output in combination with the physical mechanism. In some embodiments, the non-stationary feature sequence determined as caused by device aging in step S10 is obtained, as well as the corresponding statistical feature type, associated components, and change trend; For example, the vibration mean-time series (#1 bearing X direction): non-stationary, trend term β = 0.015 (> 0), P = 0.02; Pressure baseline-time series (air outlet): non-stationary, trend term β = -5.2 (<0), P = 0.03; Extract the initial health status data of the centrifugal fan as the comparison benchmark data, and pre-process the comparison benchmark data; Wherein, the comparison benchmark data includes statistical features corresponding to non-stationary features; For example: the vibration mean baseline of bearing #1 is 2.5g, the air outlet pressure baseline is 1200Pa, and the bearing seat temperature baseline is 45°C; The process of comparative analysis based on the time dimension is as follows: For each non-stationary feature, calculate the relative deviation between the current value and the baseline value; The relative deviation is: within the time window, the absolute value of the difference between the current value and the reference value corresponding to each data point is calculated, and the ratio of the absolute value of the difference to the reference value is calculated; The difference between the relative deviation and the preset relative deviation threshold is calculated, and the ratio of the difference to the preset relative deviation threshold is calculated to obtain the severity of the deviation from the baseline. The ratio of the number of data points with positive severity of the deviation from the baseline to the total number of data points is calculated as the proportion of the number of data points exceeding the standard; Within the time window, the relative deviations corresponding to adjacent data points are calculated for difference, and the data points with positive differences are extracted. The ratio of the number of data points with positive differences to the total number of data points is calculated as the proportion of trend growth. The product of the percentage of deviation exceeding the standard and the percentage of trend growth is calculated to output the aging trend index; It should be noted that the percentage of deviations exceeding the standard reflects the extent to which a feature deviates from the baseline value within the time window and exceeds the preset threshold. A higher percentage indicates that the feature is in a significantly abnormal state at more times and the range of deviation from the healthy state of the equipment is wider. The percentage of trend growth reflects the direction of relative deviation evolution over time. A higher percentage indicates that the proportion of deviations at adjacent moments showing an increasing trend is greater, and the degree of feature abnormality is continuously worsening, which is consistent with the physical law of gradual performance degradation during equipment aging. The coupling strength between the abnormal range and the trend of continuous deterioration is characterized from the static and dynamic dimensions respectively by using the two indicators of the proportion of the number of deviations exceeding the standard and the proportion of the number of trend growth. If the aging trend index is greater than the aging trend index limit, the corresponding non-stationary feature is determined to be a suspicious aging feature, and the components corresponding to the suspicious aging feature are enumerated into a set of suspicious aging components based on the mapping relationship between features and components; For example, the X-direction vibration mean corresponds to the drive end bearing, and the outlet pressure baseline corresponds to the impeller; If the aging trend index is less than or equal to the aging trend index limit, the corresponding non-stationary feature is determined to be a non-suspicious aging feature; Among them, the role of the aging trend index is: first, by multiplying the proportion of deviations exceeding the standard and the proportion of trend growth, the abnormal range of non-stationary features and the deterioration trend are coupled into a single quantitative indicator, which intuitively reflects the severity of the impact of aging on the features; second, by comparing with the preset aging trend index limit, it accurately determines which non-stationary features are suspicious aging features, providing a basis for subsequent locking of related components. If the index exceeds the limit, the corresponding feature is marked as suspicious, and then the set of suspicious aging components is listed through the mapping relationship between features and components, reducing the omission of real aging signals or misjudgment of irrelevant features; third, the index integrates information from both static and dynamic dimensions, ensuring the identification of features that significantly deviate from the healthy state and capturing the trend of continued deterioration of abnormalities, which is in line with the physical law of gradual performance decline during equipment aging. It provides a reliable quantitative judgment standard for time dimension comparison and analysis, and improves the scientificity and accuracy of aging feature identification. The process of comparative analysis for spatial dimensions is as follows: For any suspected aging component in the set of suspected aging components; The Pearson correlation coefficient is used to quantify the correlation strength of different sensor data from the same component in the aging equipment, and to determine whether the non-stationary characteristics are caused by the actual aging of the component, rather than due to other reasons such as sensor failure. For any combination of two different sensors in the same component, calculate the Pearson correlation coefficient in the healthy state, average the Pearson correlation coefficients corresponding to all combinations to obtain the average correlation coefficient; Count the number of sensors with suspicious aging characteristics in the component and compare it with the total number of sensors in the component to obtain trend consistency; The average correlation coefficient and the trend consistency were multiplied to obtain the correlation consistency index; Among them, the role of the correlation consistency index is: first, by integrating the average correlation coefficient in the healthy state with the current trend consistency, it quantitatively judges whether the non-stationary characteristics are caused by the real aging of the component, rather than other factors such as sensor failure and interference, and provides the core basis for distinguishing real aging from false anomalies; second, the higher the value, the higher the matching degree of the current abnormal characteristics with the physical correlation of the component in the healthy state, and the simultaneous abnormal trend of multiple sensors indicates that the feature credibility is stronger, which can screen out reliable suspicious aging characteristics, eliminate misjudgments caused by single sensor failure, and improve the accuracy of spatial dimension analysis; third, only the components whose correlation consistency index exceeds the limit value, their corresponding suspicious aging characteristics are judged to be reliable, and then combined with the physical mechanism verification output to obtain the final aging target components, ensuring the logical rigor of the deduction from suspicious components to confirmed aging components, and providing important support for accurately locating aging components and reducing invalid analysis; It should be noted that the correlation consistency index is the coupling of the baseline correlation in the healthy state and the current abnormal trend. The core is whether the current non-stationary feature is a true reflection of the abnormal state of the component itself. The higher the value, the higher the credibility of the non-stationary feature, providing a key basis for outputting aging components. If the correlation consistency index is less than or equal to the correlation consistency index limit, the suspicious aging feature corresponding to the component is determined to be unreliable; If the correlation consistency index is greater than the correlation consistency index limit, it is determined that the suspicious aging feature corresponding to the component is reliable; If the suspicious aging characteristics corresponding to the component are reliable, then combined with the physical mechanism verification, the suspicious aging component that meets the physical mechanism is output as the target component that has aged; It should be noted that the physical mechanism is a clear physical manifestation well known to those skilled in the art. For example, the physical law of bearing aging is "wear causes the gap between the rolling element and the inner and outer rings to increase → the vibration mean value increases and the high-frequency vibration (bearing characteristic frequency) energy increases → increased friction causes the temperature to rise." If the suspicious feature is "the vibration mean value and temperature increase simultaneously, and the proportion of high-frequency vibration energy increases," then it conforms to the bearing aging law. Impeller aging: The physical law is "corrosion / dust accumulation leads to impeller imbalance → decreased wind pressure (pressure baseline), increased flow fluctuations → decreased vibration energy proportion of blade pass frequency". If the suspicious feature is "decreased pressure baseline + increased flow fluctuations", it conforms to the impeller aging law; Motor aging: The physical law is "winding aging → increased current fluctuations, higher temperatures → decreased output efficiency (weakened correlation with load)." If the suspicious characteristic is "current fluctuations + simultaneous temperature increases," it conforms to the law of motor aging. The common goal of step S20 and step 10 is to capture the aging status of the equipment and provide a basis for fault trend prediction. Step S10 focuses on identifying whether data non-stationarity is caused by equipment aging. Step S20 builds on this by further accurately locating specific aging components and determining whether they have entered the aging stage. This is a deeper analysis of the equipment status from the existence of aging phenomena to the identification of aging locations and severity. Step S30: Automatically switch to a dynamically updated reinforcement learning model strategy to predict the failure trend of the aging target component; The dynamically updated reinforcement learning model strategy includes the following core mechanisms: dynamic sliding data window and healthy forgetting factor setting mechanism; By incorporating newly acquired multi-source data in real time and dynamically updating model parameters, the prediction lag caused by static models trained on historical health data can be effectively reduced, improving the accuracy of predicting the evolution trend of equipment failures during the accelerated aging phase. For the target component that has aged and is output in step S20, a reinforcement learning model framework is constructed, wherein the framework includes a state space, an action space, and a reward function; State space: takes the multi-source features of the target component as input, including: real-time statistical features: current state parameters such as vibration mean, temperature mean, pressure baseline, etc. within the time window; Trend characteristics: aging trend index, relative deviation growth rate, and correlation consistency index; Physical constraint features: correlation parameters based on physical mechanisms; Action space: defines the dynamic adjustment strategy of model parameters, including: dynamic sliding window size adjustment, for example, from 2 hours to 24 hours to adapt to data changes during the accelerated aging phase; Weight allocation of multi-source features, for example, increasing the weight of high-frequency vibration features and weakening noise interference features; Adjustment of the forecast step size, for example, shortening the forecast from 7 days to 3 days to improve the accuracy of short-term forecasts; Reward function: The goal is to minimize the prediction error; The formula is: ; The relative prediction error is the deviation ratio between the predicted value and the actual value, and the prediction lag time is the difference between the model-predicted fault occurrence time and the actual occurrence time. a and b are weight coefficients (calibrated based on historical fault data, such as a=0.6 and b=0.4), which reward "high-precision, low-lag" prediction results. First, the real-time update mechanism of the dynamic sliding data window is: Dynamically adjust the data window according to the aging stage of the target component to ensure that the model captures the latest status; Based on the historical aging data of the target component, the initial data window length is set; Optionally, 30 days of data can be taken for bearing aging and 60 days of data can be taken for impeller corrosion to match the aging evolution cycle of different components; When newly collected multi-source data is incorporated into the time series database, if the average growth rate of the relative deviation of the target component within the initial data window length exceeds the preset growth rate threshold (2%), the data window automatically slides forward by one time unit (e.g., one hour), discarding the oldest data and retaining the latest data; Among them, the core role of the average growth rate of relative deviation is to quantify the overall deterioration rate of the non-stationary characteristics of the target component within the initial data window; The second specific setting mechanism of the healthy forgetting factor is: The calculation formula of healthy forgetting factor is: ;in, is the healthy forgetting factor, k is the attenuation coefficient, generally ranging from 0.5 to 1.0, and m is the aging trend index; For example, when the aging trend index is 0.8, the attenuation coefficient is 0.6, and the health forgetting factor is 0.61, then the weight of the health data is only 61% of the current data; The data weight at the nth moment in the sliding data window is , N is the current moment, so that the weight of the newly collected multi-source data is higher than the multi-source data collected previously, and the ability of the fault trend prediction model to capture the aging trend is enhanced; For example, if the current time is the 10th hour (N=10), and the window contains data from the 6th to the 10th hour (a total of 5 time points), then n can take the values ​​6, 7, 8, 9, and 10; The 10th hour (n=10) is the current data, with a weight of 1; The data weight for the 9th hour (n=9) is 0.61; The data weight for the 6th hour (n=6) is 0.13 (the lowest weight); Third, the dynamic training and parameter update process of the reinforcement learning model is as follows: Mapping multi-source data in a dynamic sliding data window into state space parameters and inputting them into the reinforcement learning agent; The agent selects the optimal strategy from the action space based on the current state and outputs the fault trend prediction results, including: Fault occurrence time window (e.g., "bearing jam may occur within the next 15 days"); Fault severity levels, including mild aging, moderate aging, and severe aging; Key feature evolution curves, for example, the predicted fitting curve of the vibration mean over time; Compare the actual collected new data with the predicted results and calculate the reward value; If the prediction error is <5% and there is no lag, a positive reward (>) is given; If the prediction error is >10% or the lag exceeds 2 days, a negative reward (<) is given; Based on the reward value, the agent's policy network parameters are updated through the temporal difference algorithm (TD algorithm), so that the model prioritizes action strategies corresponding to high rewards, such as optimized window size and feature weights; Compare the prediction results with the physical mechanism of the target component. If they are consistent with the physical mechanism, the prediction is confirmed to be valid. If they are not consistent with the physical mechanism, the model is retrained and the weight of the physical constraint features is increased. According to the predicted fault severity level, the corresponding warning is output: Mild aging: Push trend warning and recommend increasing monitoring frequency; Moderate aging: Push maintenance warnings and prompt to prepare spare parts; Severe aging: Push emergency warning and recommend shutdown for maintenance to prevent the fault from expanding; After each round of prediction, the newly collected actual data is incorporated into the dynamic window. Through continuous iteration, the model is constantly adapted to the aging evolution trend of the target component, thereby optimizing the lag problem of the static model in "non-stationary" data, and ultimately achieving accurate prediction of the time and severity of fault occurrence; This embodiment has at least the following effects: through multi-source data monitoring and statistical characteristic analysis, combined with the ADF unit root test, it identifies data non-stationary changes caused by equipment aging, distinguishes non-stationarity driven by true aging from random interference and sensor noise, reduces the interference of invalid data in subsequent analysis, lays a reliable preliminary judgment foundation for the entire fault trend prediction process, and ensures that in-depth analysis is carried out only on characteristics related to equipment aging; Based on time and space dual-dimensional comparison analysis, suspicious aging features are screened by calculating relative deviation and aging trend index. The reliability of the features is verified by combining the correlation consistency index. Finally, the target aging components are locked through physical verification. This effectively eliminates non-aging factors such as sensor failure and environmental interference, improves the accuracy of aging component identification, clarifies the specific targets of fault prediction, reduces resource waste caused by blind maintenance, and provides a precise basis for targeted maintenance. A dynamically updated reinforcement learning model strategy is adopted for aging target components. New data is incorporated in real time through a dynamic sliding data window, and the health forgetting factor weakens the impact of outdated health data, so that the model can adapt to the non-stationary evolution trend of component aging. At the same time, through dynamic training and parameter updating of reinforcement learning, the prediction lag problem of traditional static models for non-stationary data is optimized, and the prediction accuracy of fault occurrence time and severity is improved, providing scientific decision-making support for the preventive maintenance of centrifugal fans and reducing the risk of fault expansion.

[0010] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A centrifugal fan fault trend prediction method based on multi-source information fusion, characterized by: The following steps are involved: During the operation of the centrifugal fan, by monitoring multi-source data and analyzing statistical characteristics, the changing trend of data distribution is captured, and it is identified whether the centrifuge has caused non-stationary data changes due to equipment aging; If the non-stationary changes are caused by equipment aging, the initial health status data of the equipment is used as the baseline value, and a comparative analysis of the two dimensions of time and space is performed; The process of the dual-dimensional comparison analysis of time and space is as follows: first, for the features corresponding to non-stationary changes, the relative deviation of each data point is calculated within a preset time window. By analyzing the relative deviation, an aging trend index is outputted. The aging trend index is used to extract suspicious aging features. Then, for the suspicious aging components mapped by the suspicious aging features, the correlation consistency index of the sensor in the healthy state is calculated. The correlation consistency index is used to identify the reliability of the suspicious aging components corresponding to the suspicious aging features. If reliable, then combine the physical mechanism to verify the target component whose output has aged; For target components that are aging, a dynamic updated reinforcement learning model strategy is automatically switched to predict fault trends. The reinforcement learning model includes setting a dynamic sliding data window and introducing a healthy forgetting factor.

2. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 1, characterized in that: The process of identifying whether the centrifuge's data changes due to equipment aging are as follows: The ADF unit root test method is used to analyze the multi-source data time series and identify non-stationary changes; A regression model with ADF test is established for the statistical characteristic sequence of multi-source data of centrifugal fans. The parameters of the regression model are estimated by the least squares method, and the ADF statistic is calculated; Compare the ADF statistic with a predetermined significance level, which is determined by the theoretical distribution of the ADF test. If the ADF statistic is greater than or equal to the critical value, the series is determined to be non-stationary; If the sequence is judged to be non-stationary and the trend term is significant, that is, the P value is less than 0.05 through the t-test, it is judged to be a non-stationary change caused by the equipment.

3. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 1, characterized in that: The process of obtaining the suspected aged components is as follows: In the time window, the absolute value of the difference between the current value and the reference value corresponding to each data point is calculated, and the ratio of the absolute value of the difference to the reference value is calculated as the relative deviation; By analyzing the relative deviation, the proportion of deviation exceeding the standard and the proportion of trend growth are calculated, and the product is processed to output the aging trend index; If the aging trend index exceeds the standard, the corresponding non-stationary feature is determined to be a suspicious aging feature, and the components corresponding to the suspicious aging feature are listed through the mapping relationship between features and components.

4. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 3, characterized in that: The calculation process of the percentage of deviations exceeding the standard is as follows: The relative deviation is compared with the deviation rate of the preset relative deviation threshold to obtain the severity of the deviation from the baseline. The ratio of the number of data points with positive deviation baseline severity to the total number of points is calculated as the proportion of the number of deviations exceeding the standard.

5. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 3, characterized in that: The calculation process of the trend growth ratio is as follows: Calculate the difference between the relative deviations corresponding to adjacent data points, extract the data points with positive differences, and calculate the ratio of the number of data points with positive differences to the total number of data points as the proportion of trend growth.

6. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 1, characterized in that: The process of identifying the reliability of the suspicious aging features corresponding to the suspicious aging components through the correlation consistency index is as follows: Obtain the average correlation coefficient and trend consistency, and perform product calculation to obtain the association consistency index; If the correlation consistency index is greater than the exceedance, it is determined that the suspicious aging feature corresponding to the component is reliable.

7. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 6, characterized in that: The calculation process of the average correlation coefficient is: For any combination of two different sensors in the same suspected aging component, the Pearson correlation coefficient in a healthy state is calculated, and the Pearson correlation coefficients corresponding to all combinations are averaged to obtain the average correlation coefficient.

8. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 6, characterized in that: The calculation process of the trend consistency is: The number of sensors with suspicious aging characteristics in the suspicious aging component is counted and compared with the total number of sensors in the component to obtain the trend consistency.

9. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 1, characterized in that: The dynamically updated reinforcement learning model strategy includes: For target components that are experiencing aging, a reinforcement learning model framework is constructed. The framework includes: state space, action space, and reward function; Based on the historical aging data of the target component, the initial data window length is set; When newly collected multi-source data is incorporated into the time series database, if the average growth rate of the relative deviation of the target component within the initial data window length exceeds the standard, the data window automatically slides forward by one time unit to form a dynamic sliding data window, eliminating the oldest data and retaining the latest data; The healthy forgetting factor is introduced. The calculation formula of the healthy forgetting factor is: ;in, is the healthy forgetting factor, k is the attenuation coefficient, and m is the aging trend index; The data weight at the nth moment in the sliding data window is , N is the current moment, so that the weight of the newly collected multi-source data is higher than the multi-source data collected before, and the model's ability to capture the aging trend is enhanced.

10. The method for predicting centrifugal fan fault trends based on multi-source information fusion according to claim 9, characterized in that: The dynamically updated reinforcement learning model strategy also includes: dynamic training and parameter updating of the reinforcement learning model and output of fault severity levels; Mapping multi-source data in a dynamic sliding data window into state space parameters and inputting them into the reinforcement learning agent; The agent selects the optimal strategy from the action space based on the current state and outputs the fault trend prediction results, including: Fault occurrence time window and fault severity level; Compare the actual collected new data with the predicted results and calculate the reward value; Based on the reward value, the agent's policy network parameters are updated through the temporal difference algorithm; The prediction results are compared with the physical mechanism of the target component. If they are consistent with the physical mechanism, the prediction is confirmed to be valid. If they are not consistent with the physical mechanism, the model is triggered to retrain and the weight of the physical constraint features is increased.

Images