Fault intelligent diagnosis and real-time alarm control method for direct drinking water equipment

Abstract

The invention provides a direct drinking water equipment fault intelligent diagnosis and real-time alarm control method, and relates to the technical field of intelligent monitoring and fault diagnosis, and the method comprises the steps: collecting pressure, flow, conductivity, turbidity and temperature parameters through a multi-source sensor, and transmitting the parameters to a central processing unit; the method comprises the following steps: preprocessing data by adopting dynamic wavelet transform, optimizing a mother wavelet function and a decomposition scale based on a time-frequency resolution index, and filling an abnormal value in combination with sliding window detection and linear interpolation to generate a standardized data set; extracting significant frequency band characteristics of a time domain mean value, a variance and frequency domain energy, and performing dynamic dimension reduction and normalization through principal component analysis; inputting the feature vector into a support vector machine and random forest hybrid model, and outputting a fault type and probability; dynamically adjusting an alarm threshold according to the probability, introducing a water quality anomaly index to evaluate a conductivity and turbidity collaborative risk, triggering graded alarm and pushing the graded alarm to a terminal; model parameters are updated by adopting an incremental learning algorithm, and a response strategy is optimized in combination with equipment load and environment temperature.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and fault diagnosis, and particularly to an intelligent fault diagnosis and real-time alarm control method for direct drinking water equipment. Background Art

[0002] The fault diagnosis and alarm technology of direct drinking water equipment is the core link to ensure the safety of drinking water, but the existing solutions have significant defects at multiple key levels. In the data acquisition and preprocessing stage, most systems rely on a single sensor or a fixed filtering algorithm (such as moving average), which is difficult to effectively process non-stationary signals under complex working conditions, resulting in incomplete noise filtering or loss of effective high-frequency features. For example, sudden interference of the conductivity sensor may be misjudged as water quality abnormality, and the traditional filtering method lacks the ability to dynamically adjust the scale and cannot accurately distinguish noise from real fault signals.

[0003] In terms of feature extraction, the existing technologies are mostly limited to time-domain statistics (such as mean, variance) or simple frequency-domain analysis, ignoring the multi-scale time-frequency characteristics of signals. Early weak features (such as energy changes in specific frequency bands) of progressive faults such as filter element blockage are easily missed, affecting the timeliness of fault warning. In addition, feature dimension reduction mostly uses principal component analysis with a fixed threshold, without considering the dynamic changes of the equipment operating state, resulting in insufficient characterization ability of the reduced-dimensional features for compound faults.

[0004] Fault diagnosis models generally adopt a single algorithm (such as support vector machine or random forest), with insufficient classification accuracy for multi-fault concurrent scenarios, and the model parameters are fixed, unable to adapt to the data distribution shift caused by equipment aging or water quality fluctuations. The alarm strategy relies on a static probability threshold (such as a fixed threshold of 0.9), without integrating variables such as real-time load and environmental temperature, resulting in an increase in false alarm rate under high load or a response delay for sudden water quality deterioration. Water quality monitoring analyzes parameters such as conductivity and turbidity in isolation, lacking multi-parameter collaborative evaluation indicators and being difficult to identify correlation risks such as filter membrane damage. Summary of the Invention

[0005] In order to solve the technical problems of poor dynamic adaptability in data preprocessing, single feature extraction dimension, weak generalization ability of fault diagnosis models, static solidification of alarm thresholds, and lack of multi-parameter collaborative evaluation of water quality in the prior art, the present invention provides an intelligent fault diagnosis and real-time alarm control method for direct drinking water equipment.

[0006] The technical solution provided by the present invention is as follows: An intelligent fault diagnosis and real-time alarm control method for direct drinking water equipment provided by the present invention includes: S1. Collect the operating parameters of the direct drinking water equipment through multi-source sensors, including pressure value, flow value, conductivity, pH value, turbidity, and temperature value, and transmit the data to the central processor; S2. Preprocess the collected data, including data normalization, noise filtering, and outlier removal, to generate a standardized data set; S3. Based on the standardized data set, extract time-domain features and frequency-domain features through a dynamic feature extraction algorithm to generate a multi-dimensional feature vector; S4. Input the multi-dimensional feature vector into a pre-trained fault diagnosis model to output the fault type and fault probability; the fault diagnosis model is constructed based on a hybrid model of multi-class support vector machine and random forest; S5. According to the comparison result between the fault probability and a preset threshold, trigger a real-time alarm control strategy: if the fault probability exceeds the first threshold (0.9), activate a first-level alarm and execute the equipment shutdown instruction; if the fault probability is between the second threshold (0.7) and the first threshold, activate a second-level alarm and reduce the equipment operating power; if the fault probability is lower than the second threshold, continuously monitor and update the fault diagnosis result; S6. Push the alarm information, fault type, and processing instructions to the user terminal and the cloud server through a wireless communication module.

[0007] Further, the S2 specifically includes: S201. Use wavelet transform to filter the noise of the original data, and realize signal reconstruction by selecting the optimal mother wavelet function and decomposition scale; S202. Standardize the filtered data to eliminate the dimension difference; S203. Detect outliers through the sliding window method. If the standard deviation of three consecutive data points exceeds twice the historical mean, mark it as abnormal data; S204. Interpolate and fill the abnormal data, and the filling method is linear interpolation of the valid data points before and after; S205. Verify the data integrity. If the missing data exceeds 10% of the total collected amount, trigger the sensor self-check program.

[0008] Further, the selection of the mother wavelet function in the S201 includes: S2011. Calculate the time-frequency resolution index of different mother wavelets, and the formula is: ; where, is the time window width, is the frequency band width; S2012. Select the mother wavelet with the largest TFRI value for signal decomposition; S2013. Verify the signal energy retention rate after decomposition. If the retention rate is lower than 95%, re-select the mother wavelet; S2014. Record the mother wavelet selection log for subsequent fault analysis and traceability.

[0009] Furthermore, the dynamic feature extraction algorithm in step S3 includes: S301. Calculate time-domain features, including mean, variance, peak-to-peak value, and root mean square value; S302. Perform a fast Fourier transform on the signal and extract the amplitudes of the top 5 frequency bands with the largest frequency-domain energy ratio as frequency-domain features; S303. Reduce the dimensions of the time-domain and frequency-domain features through principal component analysis and retain the principal components with a cumulative contribution rate exceeding 85%; S304. After generating the multi-dimensional feature vector, perform feature vector normalization processing; S305. Verify the stability of the feature vector. If the variance exceeds the preset threshold, re-extract the features.

[0010] Furthermore, the calculation method of the fault probability in S5 includes: S501. Calculate the fault probability according to the deviation degree between the current feature vector and the normal state feature vector. The formula is: ; where is the feature deviation degree, and the calculation formula is: ; is the total number of features, is the coefficient calibrated through the training set, is the real-time measurement value of the th dimension of the current feature vector; and are the mean and standard deviation of the th feature in the normal state respectively; S502. Dynamically adjust the first threshold and the second threshold according to the equipment operation time. For every 100 hours increase in the operation time, the threshold is reduced by 0.02; S503. If the calculated result of the fault probability exceeds 0.8 for 5 consecutive times but does not reach the alarm threshold, trigger a manual re-inspection request; S504. Record all the calculated results of the fault probability and the adjustment log to form a historical database.

[0011] Furthermore, the calibration method of the coefficient in S501 includes: S5011. Construct a training set based on historical fault data, covering the normal state and 5 typical fault states; S5012. Use the gradient descent method to optimize , and the objective function is to minimize the mean square error between the model output and the true label; S5013. Verify the coefficient stability through the ten-fold cross-validation method. If the variance exceeds 0.1, re-optimize it. S5014. Introduce the L2 regularization term to prevent overfitting. The objective function is modified as: ; where, is the loss function value, is the total number of training samples, is the predicted failure probability of the th sample, is the th sample's true failure label, 0 or 1, is the regularization coefficient; S5015. Encrypt and store the calibrated coefficients in the security module.

[0012] Furthermore, the trigger conditions for the first-level alarm in step S5 further include: S511. Trigger the first-level alarm when the failure probability meets the following composite conditions: ; where, is the current failure probability, is the time required for the failure probability to rise from 0.5 to 0.9; S512. If the device is in a high-load operation state, lower the trigger threshold of the first-level alarm to 0.85; S513. Detect whether the ambient temperature exceeds 40°C. If it does, additionally shorten to 8 seconds; S514. Record the device operation status and environmental parameters at the time of alarm trigger, and generate an alarm event report.

[0013] Furthermore, the fault location method in S6 includes: S601. Based on the fault type, match the preset fault-component mapping table to determine the possible faulty components; S602. Screen the target component with the highest correlation through the component correlation matrix, and the correlation matrix is calculated by the product of the feature weight and the component sensitivity; S603. Combine the device maintenance records and preferentially screen the components with high recent fault frequencies; S604. If the confidence level of the component with the highest correlation is lower than 80%, start the multi-component joint diagnosis process; S605. Generate a visualization chart of the fault location result and mark the top three components in terms of correlation.

[0014] Furthermore, the dynamic update of the fault diagnosis model in S4 includes: S401. Every interval of time Collect newly added fault data and label the fault type; S402. Update the parameters of the hybrid model using the incremental learning algorithm; S403. If the model accuracy drops by more than 5%, trigger a full retraining; S404. Record the timestamp and version number of each model update for the user terminal to trace; S405. Verify the recall rate of the updated model on the test set. If it is lower than 90%, roll back to the previous version.

[0015] Furthermore, the control strategy for the secondary alarm in S5 further includes: S521. Calculate the water quality anomaly index based on real-time water quality parameters. The formula is: ; Wherein, and are proportionality coefficients and satisfy + , is the real-time conductivity, with the unit of ; is the real-time turbidity, with the unit of NTU; is the conductivity reference value under normal conditions, is the turbidity threshold; S522. If the WQAI exceeds 0.8, force an upgrade to a primary alarm; S523. Dynamically adjust the WQAI threshold according to the water quality safety level preset by the user.

[0016] The beneficial effects brought by the technical solution provided by the present invention at least include: (1) In the present invention, through the dynamic wavelet transform and adaptive mother wavelet selection technology, the accuracy and adaptability of data preprocessing are significantly improved. For the characteristics of non-stationary signals, the time-frequency resolution index (TFRI) is used to optimize the mother wavelet function and decomposition scale, effectively distinguishing noise from real fault signals and avoiding the problems of feature loss or noise residue caused by traditional fixed filtering algorithms. Combining sliding window anomaly detection and linear interpolation filling ensures data integrity while reducing the misjudgment rate. This technical means solves the problem of poor dynamic adaptability of data preprocessing in the background technology and provides a standardized data set with high signal-to-noise ratio for subsequent feature extraction and fault diagnosis.

[0017] (2) In the present invention, through the integration of multi-dimensional features in the time domain and frequency domain and dynamic dimensionality reduction technology, the comprehensive mining and efficient characterization of fault features are realized. Based on the fast Fourier transform, the frequency bands with significant frequency domain energy are extracted, and combined with principal component analysis, the principal components with a cumulative contribution rate exceeding 85% are dynamically retained, overcoming the defects of single feature dimension and information redundancy in traditional methods. This technical means can capture the early weak features of faults such as filter element blockage and water quality deterioration, solves the problem of single feature extraction dimension in the background technology, and significantly improves the timeliness and classification accuracy of fault warning.

[0018] (3) In the present invention, through the hybrid fault diagnosis model and dynamic threshold alarm strategy, the adaptability of the system to complex fault scenarios is enhanced. The hybrid architecture of support vector machine and random forest is adopted, and the model parameters are dynamically updated in combination with the incremental learning algorithm to effectively cope with equipment aging and data distribution shift. The fault probability calculation model and water quality anomaly index (WQAI) are introduced, and the alarm threshold is adjusted in real time by fusing variables such as equipment load and environmental temperature, solving the problems of false alarms and missed alarms caused by static thresholds. This technical means realizes the accurate judgment of multi-fault concurrency and the associated changes in water quality risks, and comprehensively improves the reliability and intelligence level of the system. Brief Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 It is a schematic flowchart of a method for intelligent fault diagnosis and real-time alarm control of a direct drinking water equipment provided by an embodiment of the present invention; Figure 2 It is a schematic flowchart of preprocessing in a method for intelligent fault diagnosis and real-time alarm control of a direct drinking water equipment provided by an embodiment of the present invention; Figure 3 It is a schematic flowchart of a dynamic feature extraction algorithm in a method for intelligent fault diagnosis and real-time alarm control of a direct drinking water equipment provided by an embodiment of the present invention; Figure 4 It is a schematic flowchart of a fault location method in a method for intelligent fault diagnosis and real-time alarm control of a direct drinking water equipment provided by an embodiment of the present invention. Detailed Description of the Embodiments

[0021] The following will describe the technical solutions in the present invention with reference to the drawings.

[0022] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to give examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of the word "example" is intended to present concepts in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two can be selected.

[0023] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same. "(of)", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same.

[0024] In the embodiments of the present invention, sometimes subscripts such as may be miswritten as non-subscript forms such as W1. When the difference is not emphasized, the meanings they express are the same.

[0025] To make the technical problems to be solved, technical solutions and advantages of the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0026] Referring to the attached Figure 1 figures, a schematic flow chart of a method for intelligent diagnosis and real-time alarm control of faults in a direct drinking water device provided by an embodiment of the present invention is shown.

[0027] The embodiments of the present invention provide a method for intelligent diagnosis and real-time alarm control of faults in a direct drinking water device. The processing flow may include the following steps: S1. Collect the operating parameters of the direct drinking water device through multi-source sensors, including pressure values, flow values, conductivity, pH values, turbidity and temperature values, and transmit the data to the central processor.

[0028] It should be noted that the operating parameters of the direct drinking water device are collected through multi-source sensors, including pressure values, flow values, conductivity (Conductivity, CON), pH values, turbidity (Turbidity, TUR) and temperature values. The sensors transmit the data to the central processor at a frequency of once per second. The central processor conducts a preliminary verification of the data to ensure that the data format conforms to the preset specification. If the verification fails, a data retransmission mechanism is triggered. The collected data is stored in the buffer in the order of timestamps. The buffer capacity is set to 1000 pieces of data. When the data volume reaches the capacity limit, the earliest historical data is automatically overwritten.

[0029] S2. Preprocess the collected data, including data normalization, noise filtering, and outlier removal, to generate a standardized data set.

[0030] In a possible implementation, as Figure 2 shown, the specific step flow of S2 is as follows: S201. Use wavelet transform (WT) to filter the noise of the original data and achieve signal reconstruction by selecting the optimal mother wavelet function and decomposition scale.

[0031] In a possible implementation, the selection of the mother wavelet function in S201 includes: S2011. Calculate the time-frequency resolution index (TFRI) of different mother wavelets. The formula is: ; where is the time window width, is the frequency band width. The system presets candidate mother wavelet functions including the Daubechies series (db4 to db8) and the Symlet series (sym5 to sym8). By traversing and calculating the TFRI values of each mother wavelet, select the mother wavelet corresponding to the maximum value; [[ID=2)]] S2012. Select the mother wavelet with the maximum TFRI value for signal decomposition; S2013. Verify the signal energy retention rate after decomposition. If the retention rate is lower than 95%, reselect the mother wavelet; S2014. Record the mother wavelet selection log for subsequent fault analysis and traceability.

[0032] S202. Reconstruct the filtered signal into a time-domain signal through inverse wavelet transform and perform standardization processing to eliminate the dimension difference. The calculation formula is: ; where is the standardized data value, is the data mean, is the standard deviation. The mean and standard deviation are dynamically updated based on the historical data of the last 30 days.

[0033] S203. Detect outliers through the sliding window method. The window size is set to 5 data points. If the standard deviation of 3 consecutive data points exceeds 2 times the historical mean, it is marked as abnormal data.

[0034] S204. Interpolate and fill the abnormal data. The filling method is linear interpolation of the valid data points before and after. The filling formula is: ; Among them, is the filled data value, and are respectively the values of the first valid data points before and after the abnormal data point, is the abnormal data timestamp, and are respectively the timestamps of the valid data before and after.

[0035] S205. Verify the data integrity. If the missing data exceeds 10% of the total acquisition volume, trigger the sensor self-check program, and the self-check includes sensor power supply detection and signal link test.

[0036] S3. Based on the standardized data set, extract time-domain features and frequency-domain features through the dynamic feature extraction algorithm to generate a multi-dimensional feature vector.

[0037] In a possible implementation manner, as Figure 3 shown, the dynamic feature extraction algorithm in step S3 includes: S301. Calculate the time-domain features, including mean, variance, peak-to-peak value, and root mean square value (Root Mean Square, RMS). The peak-to-peak value is the difference between the maximum value and the minimum value within the window. The RMS calculation formula is: ; Among them, is the number of data points within the window.

[0038] S302. Perform a fast Fourier transform (Fast Fourier Transform, FFT) on the signal, and extract the amplitudes of the top 5 frequency bands with the largest frequency-domain energy proportion as the frequency-domain features; S303. Perform dimensionality reduction on the time-domain and frequency-domain features through principal component analysis (Principal Component Analysis, PCA), and retain the principal components with a cumulative contribution rate exceeding 85%; S304. After generating the multi-dimensional feature vector, perform feature vector normalization processing. The normalization range is [0,1], and the formula is: ; Among them, is the normalized eigenvalue (range [0,1]); is the original eigenvalue; is the minimum value of this feature in the current feature vector; is the maximum value of this feature in the current feature vector.

[0039] S305. Verify the stability of the feature vector. If the variance exceeds the preset threshold of 0.05, re-extract the features.

[0040] S4. Input the multi-dimensional feature vector into the pre-trained fault diagnosis model, and output the fault type and fault probability. The fault diagnosis model is constructed based on a hybrid model of multi-class support vector machine (Support Vector Machine, SVM) and random forest (RandomForest, RF). The support vector machine is responsible for the initial classification, and the random forest corrects the classification result through a voting mechanism.

[0041] In a possible implementation manner, the dynamic update of the fault diagnosis model in S4 includes: S401. Collect new fault data and label the fault type every interval of time (default 24 hours); S402. Update the parameters of the hybrid model using an incremental learning algorithm; S403. If the model accuracy drops by more than 5%, trigger full-scale retraining; S404. Record the timestamp and version number of each model update for user terminal tracing; S405. Verify the recall rate of the updated model on the test set. If it is lower than 90%, roll back to the previous version.

[0042] S5. According to the comparison result between the fault probability and the preset threshold, trigger a real-time alarm control strategy: if the fault probability exceeds the first threshold (0.9), start a first-level alarm and execute the equipment shutdown instruction. If the fault probability is between the second threshold (0.7) and the first threshold, start a second-level alarm and reduce the equipment operating power. If the fault probability is lower than the second threshold, continuously monitor and update the fault diagnosis result.

[0043] In a possible implementation manner, the calculation method of the fault probability in S5 includes: S501. Calculate the fault probability according to the deviation degree between the current feature vector and the normal state feature vector. The formula is: ; where is the feature deviation degree, and the calculation formula is: ; is the total number of features, is the feature index , is the coefficient calibrated through the training set, [[ID=4,5]] is the real-time measurement value of the -th dimension of the current feature vector; and are respectively the mean and standard deviation of the -th feature in the normal state.

[0044] In a possible implementation, the calibration method of the coefficient in S501 includes: S5011. Construct a training set based on historical fault data, covering the normal state and 5 typical fault states; S5012. Use the gradient descent method to optimize , and the objective function is to minimize the mean square error between the model output and the true label; S5013. Verify the coefficient stability through the ten-fold cross-validation method. If the variance exceeds 0.1, re-optimize; S5014. Introduce the L2 regularization term to prevent overfitting, and the objective function is modified to: ; where, is the loss function value, is the total number of training samples, is the predicted fault probability of the th sample, is the th sample's true fault label, (0 represents normal, 1 represents fault), is the regularization coefficient, and the default setting is 0.01; S5015. Encrypt and store the calibrated coefficient in the security module.

[0045] S502. Dynamically adjust the first threshold and the second threshold according to the device running time. For every 100 hours increase in the running time, the threshold is reduced by 0.02; S503. If the calculated result of the fault probability exceeds 0.8 for 5 consecutive times but does not reach the alarm threshold, trigger a manual re-inspection request; S504. Record all the calculated results of the fault probability and the adjustment log to form a historical database.

[0046] In a possible implementation, the triggering conditions for the first-level alarm in step S5 further include: S511. Trigger the first-level alarm when the fault probability meets the following composite conditions: ; where, is the current fault probability, is the time required for the fault probability to rise from 0.5 to 0.9; S512. If the device is in a high-load operating state (power ≥ 90% of the rated value), then reduce the triggering threshold of the first-level alarm to 0.85; S513. Detect whether the ambient temperature exceeds 40°C. If it does, additionally shorten to 8 seconds; S514. Record the device operation status and environmental parameters when the alarm is triggered, and generate an alarm event report.

[0047] In a possible implementation, the control strategy for secondary alarms in S5 further includes: S521. Calculate the water quality anomaly index based on the real-time water quality parameters. The formula is: ; where and are proportionality coefficients and satisfy + , with default values of 0.6 and 0.4. is the real-time conductivity, with the unit of ; is the real-time turbidity, with the unit of ; is the conductivity reference value under normal conditions. is the turbidity threshold; S522. If exceeds 0.8, force an upgrade to a primary alarm; S523. Dynamically adjust the threshold according to the water quality safety level preset by the user.

[0048] S6. Push the alarm information, fault type, and processing instructions to the user terminal and the cloud server through the wireless communication module.

[0049] In a possible implementation, as shown in Figure 4 , the fault location method in S6 includes: S601. Based on the fault type, match the preset fault-component mapping table to determine the possible faulty components; S602. Screen the target component with the highest correlation through the component correlation matrix. The component correlation matrix is calculated by multiplying the feature weight and the component sensitivity. The calculation formula is: ; where is the correlation between the fault type and the component . is the weight of the fault type on the th feature. is the sensitivity of the component to the feature . is the total number of features; S603. Combine the device maintenance records and preferentially screen the components with a high recent fault frequency; S604. If the confidence level of the component with the highest correlation is lower than 80%, start the multi-component joint diagnosis process; S605. Generate a visualization chart of the fault location result, and mark the top three components in terms of correlation.

[0050] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include: (1) In the present invention, through the dynamic wavelet transform and the adaptive mother wavelet selection technology, the accuracy and adaptability of data preprocessing are significantly improved. Aiming at the characteristics of non-stationary signals, the time-frequency resolution index (TFRI) is used to optimize the mother wavelet function and the decomposition scale, effectively distinguishing noise from real fault signals and avoiding the problems of feature loss or noise residue caused by traditional fixed filtering algorithms. Combining sliding window anomaly detection and linear interpolation filling, while ensuring data integrity, the false positive rate is reduced. This technical means solves the problem of poor dynamic adaptability of data preprocessing in the background technology, and provides a standardized data set with high signal-to-noise ratio for subsequent feature extraction and fault diagnosis.

[0051] (2) In the present invention, by fusing multi-dimensional features in the time domain and frequency domain and dynamic dimensionality reduction technology, the comprehensive mining and efficient characterization of fault features are realized. Based on the fast Fourier transform, the frequency bands with significant frequency domain energy are extracted, and the principal components with a cumulative contribution rate exceeding 85% are dynamically retained in combination with principal component analysis, overcoming the defects of single feature dimension and information redundancy in traditional methods. This technical means can capture the early weak features of faults such as filter element blockage and water quality deterioration, solves the problem of single feature extraction dimension in the background technology, and significantly improves the timeliness of fault warning and the classification accuracy.

[0052] (3) In the present invention, through the hybrid fault diagnosis model and the dynamic threshold alarm strategy, the adaptability of the system to complex fault scenarios is enhanced. A hybrid architecture of support vector machine and random forest is adopted, and the model parameters are dynamically updated in combination with the incremental learning algorithm to effectively cope with equipment aging and data distribution shift. The fault probability calculation model and the water quality anomaly index (WQAI) are introduced, and the alarm threshold is adjusted in real time by fusing variables such as equipment load and environmental temperature, solving the problems of false alarms and missed alarms caused by static thresholds. This technical means realizes the accurate judgment of multi-fault concurrency and the change of water quality risk correlation, and comprehensively improves the reliability and intelligent level of the system.

[0053] The above content is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0054] The following points need to be noted: (1) The attached drawings of the embodiments of the present invention only relate to the structures involved in the embodiments of the present invention, and other structures can refer to the general design.

[0055] (2) For clarity, in the attached drawings used to describe the embodiments of the present invention, the thickness of the layer or region is enlarged or reduced, that is, these drawings are not drawn to actual scale. It can be understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element can be "directly" on or under the other element or there can be intermediate elements.

[0056] (3) Without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0057] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An intelligent fault diagnosis and real-time alarm control method for direct drinking water equipment, characterized in that, Including: S1. Collect the operating parameters of the direct drinking water equipment through multi-source sensors, including pressure value, flow value, conductivity, pH value, turbidity and temperature value, and transmit the data to the central processor; S2. Preprocess the collected data, including data normalization, noise filtering and outlier removal, to generate a standardized data set; S3. Based on the standardized data set, extract time-domain features and frequency-domain features through a dynamic feature extraction algorithm to generate a multi-dimensional feature vector; S4. Input the multi-dimensional feature vector into a pre-trained fault diagnosis model to output the fault type and fault probability; the fault diagnosis model is constructed based on a hybrid model of multi-class support vector machine and random forest; S5. According to the comparison result between the fault probability and the preset threshold, trigger a real-time alarm control strategy: if the fault probability exceeds the first threshold, start a first-level alarm and execute the equipment shutdown instruction; if the fault probability is between the second threshold and the first threshold, start a second-level alarm and reduce the equipment operating power; if the fault probability is lower than the second threshold, continuously monitor and update the fault diagnosis result; S6. Push the alarm information, fault type and processing instructions to the user terminal and the cloud server through the wireless communication module.

2. The intelligent diagnosis and real-time alarm control method for the failure of a direct drinking water device according to claim 1, characterized in that, The specific content of S2 includes: S201. Use wavelet transform to filter the noise of the original data, and realize signal reconstruction by selecting the optimal mother wavelet function and decomposition scale; S202. Standardize the filtered data to eliminate the dimension difference; S203. Detect outliers by the sliding window method. If the standard deviation of 3 consecutive data points exceeds 2 times the historical mean, it is marked as abnormal data; S204. Interpolate and fill the abnormal data, and the filling method is linear interpolation of the front and back valid data points; S205. Verify the data integrity. If the missing data exceeds 10% of the total collected amount, trigger the sensor self-check program.

3. An intelligent diagnosis and real-time alarm control method for the failure of a direct drinking water device according to claim 2, characterized in that, The selection of the mother wavelet function in S201 includes: S2011. Calculate the time-frequency resolution index of different mother wavelets, and the formula is: ； wherein, is the time window width, is the frequency band width; S2012. Select the mother wavelet with the largest TFRI value for signal decomposition; S2013. Verify the signal energy retention rate after decomposition. If the retention rate is lower than 95%, reselect the mother wavelet; S2014. Record the mother wavelet selection log for subsequent fault analysis and traceability.

4. An intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 1, characterized in that The dynamic feature extraction algorithm in step S3 includes: S301. Calculate time-domain features, including mean value, variance, peak-to-peak value and root mean square value; S302. Perform a fast Fourier transform on the signal, and extract the amplitudes of the top 5 frequency bands with the largest frequency-domain energy ratio as frequency-domain features; S303. Perform dimensionality reduction on the time-domain and frequency-domain features through principal component analysis, and retain the principal components with a cumulative contribution rate exceeding 85%; S304. After generating the multi-dimensional feature vector, perform normalization processing on the feature vector; S305. Verify the stability of the feature vector. If the variance exceeds the preset threshold, re-extract the features.

5. An intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 1, characterized in that The calculation method of the fault probability in S5 includes: S501. Calculate the fault probability according to the deviation degree between the current feature vector and the normal state feature vector, and the formula is: ； Among them, is the feature deviation degree, and the calculation formula is: ； is the total number of features, is the coefficient calibrated by the training set, is the real-time measurement value of the -th dimension of the current feature vector; and are the mean and standard deviation of the -th feature in the normal state, respectively; S502. Dynamically adjust the first threshold and the second threshold according to the device running time. For every 100 - hour increase in the running time, the threshold decreases by 0.02; S503. If the calculated result of the failure probability exceeds 0.8 for 5 consecutive times but does not reach the alarm threshold, trigger a manual re - inspection request; S504. Record all the calculated results of the failure probability and the adjustment log to form a historical database.

6. The intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 5, characterized in that, The calibration method for the coefficient in S501 includes: S5011. Construct a training set based on historical failure data, covering the normal state and 5 typical failure states; S5012. Optimize using the gradient descent method , and the objective function is to minimize the mean squared error between the model output and the true label; S5013. Verify the coefficient stability through the ten - fold cross - validation method. If the variance exceeds 0.1, re - optimize; S5014. Introduce the L2 regularization term to prevent overfitting, and the objective function is modified as: ； Among them, is the loss function value, is the total number of training samples, is the predicted failure probability of the th sample, is the true failure label of the [[ID=,14]]th sample, 0 or 1, is the regularization coefficient; S5015. Encrypt and store the calibrated coefficients in the security module.

7. An intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 1, characterized in that The triggering condition of the first - level alarm in step S5 further includes: S511. Trigger the first - level alarm when the failure probability meets the following composite conditions: ； Among them, is the current failure probability, is the time required for the failure probability to rise from 0.5 to 0.9; S512. If the device is in a high - load running state, reduce the triggering threshold of the first - level alarm to 0.85; S513. Detect whether the ambient temperature exceeds 40°C. If it does, shorten it additionally by up to 8 seconds; S514. Record the device running state and environmental parameters at the time of alarm triggering, and generate an alarm event report.

8. An intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 1, characterized in that, The fault location method in S6 includes: S601. Based on the fault type, match the preset fault - component mapping table to determine the possible faulty components; S602. Screen the target component with the highest correlation through the component correlation matrix, and the correlation matrix is calculated by the product of the feature weight and the component sensitivity; S603. Combine the device maintenance records and preferentially screen the components with a high recent failure frequency; S604. If the confidence level of the component with the highest correlation is lower than 80%, start the multi - component joint diagnosis process; S605. Generate a visualization chart of the fault location result and mark the top three components in terms of correlation.

9. The intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 1, characterized in that, The dynamic update of the fault diagnosis model in S4 includes: S401. At every interval of time Collect newly added fault data and label the fault type; S402. Update the parameters of the hybrid model using the incremental learning algorithm; S403. If the model accuracy drops by more than 5%, trigger a full - scale retraining; S404. Record the timestamp and version number of each model update for user - terminal traceability; S405. Verify the recall rate of the updated model on the test set. If it is lower than 90%, roll back to the previous version.

10. The intelligent fault diagnosis and real-time alarm control method for a direct drinking water device according to claim 1, characterized in that, The control strategy of the second - level alarm in S5 further includes: S521. Calculate the water quality anomaly index according to the real - time water quality parameters. The formula is: ； Among them, and are proportionality coefficients and satisfy + , is the real-time conductivity, with the unit of ; is the real-time turbidity, with the unit of ; is the conductivity reference value under normal conditions, is the turbidity threshold; S522. If exceeds 0.8, it is forced to be upgraded to a first-level alarm; S523. Dynamically adjust the threshold according to the water quality safety level preset by the user.

Images