Sewage plant blower online diagnosis system and method based on knowledge base

Abstract

The invention discloses a knowledge base-based sewage plant blower online diagnosis system and a knowledge base-based sewage plant blower online diagnosis method. The system comprises an edge-cloud collaborative architecture which is composed of an edge end deployed on an equipment site and a cloud server; the cloud server comprises a knowledge base management system, a hybrid inference engine and a model training platform; the problem of collaborative optimization of single information dimension, limited reasoning mechanism, knowledge update lag and response delay in sewage plant blower fault diagnosis is solved, and the method is realized by constructing a full-dimension knowledge base and combining an edge-cloud collaborative architecture and a rule-case-deep learning hybrid reasoning mechanism. The effects of fault diagnosis accuracy improvement, response speed acceleration and knowledge base self-evolution are achieved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology for sewage treatment equipment, and in particular relates to an online diagnostic system and method for sewage treatment plant blowers based on a knowledge base. Background Technology

[0002] As a core component of the aeration system in wastewater treatment plants, the operating status of blowers directly impacts wastewater treatment efficiency and energy consumption. Traditional blower fault diagnosis primarily relies on periodic manual inspections and offline monitoring, which suffers from slow response times and high false alarm rates. With the development of IoT technology, sensor-based online diagnostic systems are gradually being applied. Existing diagnostic systems using rule-based reasoning rely on pre-set expert rule bases, such as the ISO-10816 vibration standard, to trigger alarms when sensor data exceeds thresholds. These methods have the following drawbacks: Limited rule coverage: It can only handle known fault modes, and the failure rate for complex faults such as pneumatic-mechanical coupling failure is too high; Poor adaptability: The rule thresholds are fixed and cannot adapt to parameter drift caused by equipment aging; Weak interpretability: It lacks the ability to locate the root cause of the fault and only outputs "out of limit" alarms, requiring maintenance personnel to conduct secondary investigations.

[0003] Existing technologies employ cloud-based diagnostic platforms based on deep learning, which upload sensor data to the cloud and use models such as CNNs to identify fault modes. However, these methods have the following drawbacks: Insufficient real-time performance: The high latency introduced by data upload and cloud computing cannot meet the requirements of fault early warning with a response time of seconds; Knowledge solidification: Model training relies on historical datasets and it is difficult to integrate new expert experience or real-time cases, resulting in a sharp drop in accuracy when facing new types of faults; Black box decision-making: The model outputs the probability of faults but has no reasoning basis, making it difficult for operation and maintenance personnel to verify the credibility of the results.

[0004] Existing multi-sensor information fusion diagnostic methods, which integrate vibration, temperature, and current data and use algorithms such as DS evidence theory to comprehensively determine faults, have the following drawbacks: Lagging knowledge updates: The fault rule base relies on manual maintenance, and newly discovered fault modes require system shutdown for updates; lack of edge computing: Raw data is uploaded in full, which consumes a lot of bandwidth and cannot achieve rapid local screening.

[0005] In summary, existing blower fault diagnosis technologies based on IoT technology have the following key problems: Diagnostic information is limited: it relies on single-source data or simple fusion, and cannot collaboratively analyze multi-dimensional heterogeneous information such as mechanical vibration, aerodynamic performance, and electrical parameters. Limitations of reasoning mechanisms: rule-based reasoning struggles to handle complex nonlinear faults, while machine learning models lack interpretability, and neither implements a three-level progressive reasoning architecture. Lagging knowledge updates: Expert experience, historical cases and real-time fault data are disconnected, making it impossible to build a self-evolving knowledge management system; Response speed bottleneck: The centralized cloud processing mode causes diagnostic delays, and the edge terminal lacks intelligent initial screening capabilities, which cannot meet the needs of real-time early warning.

[0006] Therefore, a knowledge-based online diagnostic system and method for wastewater treatment plant blowers is needed to solve the above problems. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide an online diagnostic system and method for wastewater treatment plant blowers based on a knowledge base. It aims to solve the collaborative optimization problems of single information dimension, limited reasoning mechanism, lagging knowledge update and response delay in the fault diagnosis of wastewater treatment plant blowers. By constructing a full-dimensional knowledge base and combining an edge-cloud collaborative architecture with a rule-case-deep learning hybrid reasoning mechanism, the system achieves the effects of improved fault diagnosis accuracy, faster response speed and self-evolution of the knowledge base.

[0008] To achieve the above-mentioned technical effects, the technical solution adopted by the present invention is as follows: A knowledge-based online diagnostic system for wastewater treatment plant blowers includes: Edge-cloud collaborative architecture: consists of edge devices deployed at the device site and cloud servers; the cloud servers include: The knowledge base management system is used to build a full-dimensional knowledge base, including eight modules: storage device basic library, fault mode library, monitoring technology library, case experience library, maintenance strategy library, standard specification library, operating parameter library, and sensor integration library. It adopts knowledge graph structured storage. Hybrid reasoning engine: Performs a three-level progressive diagnosis on a cloud server, which combines rule-based reasoning, case-based reasoning, and deep learning reasoning, and fuses the results through DS evidence theory; The model training platform includes a self-evolution management module: optimizing the knowledge base and AI model through an anomaly-driven knowledge discovery mechanism and federated learning.

[0009] Furthermore, the cloud server connects to web visualization interfaces, mobile apps, and AR-assisted operation and maintenance application scenarios.

[0010] Preferably, the edge end is configured with: Accelerometers, thermistors, and current sensors; Preprocessing module: Performs filtering, noise reduction, and normalization processing; Feature extraction module: Calculates the effective value of vibration. , cliff 1x frequency amplitude and its proportion ,in where n is the harmonic order; Threshold judgment module: This module compares the collected data with preset thresholds in the runtime parameter library in real time, performing initial anomaly screening. Specifically, it includes: Vibration threshold judgment: When the effective value of vibration meets the following conditions... When an alert is triggered, among which This is the vibration reference value under normal operating conditions of the equipment. The safety factor is (1.2≤α≤1.5). Temperature gradient determination: When the rate of temperature change satisfies When an alert is triggered, among which The gradient threshold is (3℃ / min≤β≤8℃ / min); Composite judgment: When the current fluctuation rate Furthermore, if the temperature gradient exceeds the limit, it is judged as an emergency fault (15%≤γ≤20%).

[0011] Preferably, the hybrid inference engine comprises: Rule-based reasoning unit: Configure impeller imbalance rules: >60% and If the speed is >4.5 mm / s, the impeller is considered unbalanced. Case Reasoning Unit: The case database is retrieved based on cosine similarity, with the similarity formula as follows: ; in For real-time feature vectors, For example, the feature vector of the case; Deep learning unit: Uses CNN-LSTM model to process spectral sequences.

[0012] Preferably, the self-evolution management module includes: Dual-threshold anomaly detector: When data continuously exceeds k times the standard deviation for a duration greater than t minutes, the Isolation Forest algorithm is triggered; Rule generator: Automatically generates candidate rules for new fault modes obtained from clustering, using the format "IF condition set THEN fault type".

[0013] Preferably, the full-dimensional knowledge base is stored using knowledge graph technology and entity-relationship modeling is achieved through the Neo4j graph database. Entities include blower components, fault types, feature parameters, and solutions, while relationships include fault association, feature mapping, and solution matching.

[0014] Preferably, the online diagnostic method for wastewater treatment plant blowers based on a knowledge base includes the following steps: S1, the edge end collects vibration, temperature and current data in real time at a sampling rate of 10kHz; S2, threshold screening, performs edge-level anomaly detection: Calculate the effective value of vibration If satisfied Then mark it as an anomaly; Calculate the temperature gradient If satisfied Then mark it as an exception; When current fluctuation rate Furthermore, if the temperature gradient exceeds the limit, an emergency alarm will be triggered directly; S3 performs preprocessing on outlier data: filtering, noise reduction, and normalization; S4, Extract vibration features: kurtosis 1x frequency amplitude and its proportion ; S5, execute three-level progressive reasoning: Rule-based reasoning: Matches to a pre-defined rule base to diagnose common faults; Case-based reasoning: A solution for reusing a case database by retrieving cases using cosine similarity; Deep learning inference: Analyzing complex faults using CNN-LSTM models; S6, using the DS evidence theory to fuse the results of the three-level inference, calculates the fault confidence: ; in To have faith in the lower realm, Let m(B) be the upper confidence bound and the basic probability assignment. S7 generates a visual report containing diagnostic conclusions, supporting evidence, and maintenance recommendations.

[0015] Preferably, the rule reasoning in step S5 specifically includes: When both conditions are met >60% and When the speed is >4.5 mm / s, it is determined that the impeller is unbalanced; When temperature gradient >5℃ / min and current fluctuation rate When the value is greater than 15%, it is determined that the bearing is overheated.

[0016] Preferably, in the case reasoning of step S5, a similarity threshold is set. When the similarity between the real-time feature vector and the historical case is greater than the similarity threshold, the case solution is directly reused.

[0017] Preferably, in the deep learning inference of step S5: The CNN layer uses 3×3 convolutional kernels to extract vibration spectrum features; The LSTM layer uses 128 neurons to process time series and outputs a fault probability distribution.

[0018] Preferably, it also includes a knowledge self-evolution step: When the data continuously exceeds k times the standard deviation and lasts for more than t minutes, the Isolation Forest algorithm is triggered to detect outliers. Novel fault modes are discovered through hierarchical clustering, and candidate rules "IF condition set THEN fault type" are generated. After expert review, the knowledge base is updated and incremental training of the model is triggered.

[0019] The beneficial effects of this invention are as follows: 1. The hybrid reasoning mechanism adopted in this invention significantly improves the accuracy of fault diagnosis compared to traditional methods; the cloud-edge collaboration and edge computing approach enables second-level fault warnings, greatly reducing diagnostic latency.

[0020] 2. This invention uses knowledge graph technology to model various types of knowledge. Through the entity-relationship model, it clearly shows the relationship between equipment components, fault types, characteristic manifestations, and solutions. Compared with traditional table or text storage methods, it facilitates knowledge retrieval, reasoning, and expansion. It uses the Neo4j graph database to store the knowledge graph, supports complex semantic queries, and can quickly locate all knowledge information related to a specific fault.

[0021] 3. This invention can automatically update the knowledge base based on maintenance records submitted by maintenance personnel, and continuously adapt to equipment aging and new faults. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a flowchart illustrating the diagnostic method in an embodiment of the present invention; Figure 3 This is the waveform spectrum interface for data acquisition in this embodiment of the invention. Detailed Implementation

[0023] Example 1: like Figure 1 As shown, a knowledge-based online diagnostic system for wastewater treatment plant blowers includes: Edge-cloud collaborative architecture: consists of edge devices deployed at the device site and cloud servers; the cloud servers include: The knowledge base management system is used to build a full-dimensional knowledge base, including eight modules: storage device basic library, fault mode library, monitoring technology library, case experience library, maintenance strategy library, standard specification library, operating parameter library, and sensor integration library. It adopts knowledge graph structured storage. Hybrid reasoning engine: Performs a three-level progressive diagnosis on a cloud server, which combines rule-based reasoning, case-based reasoning, and deep learning reasoning, and fuses the results through DS evidence theory; The model training platform includes a self-evolution management module: optimizing the knowledge base and AI model through an anomaly-driven knowledge discovery mechanism and federated learning.

[0024] Furthermore, the cloud server connects to web visualization interfaces, mobile apps, and AR-assisted operation and maintenance application scenarios.

[0025] Preferably, the edge end is configured with: Accelerometers, thermistors, and current sensors; Preprocessing module: Performs filtering, noise reduction, and normalization processing; Feature extraction module: Calculates the effective value of vibration. , cliff 1x frequency amplitude and its proportion ,in where n is the harmonic order; Threshold judgment module: This module compares the collected data with preset thresholds in the runtime parameter library in real time, performing initial anomaly screening. Specifically, it includes: Vibration threshold judgment: When the effective value of vibration meets the following conditions... When an alert is triggered, among which This is the vibration reference value under normal operating conditions of the equipment. The safety factor is (1.2≤α≤1.5). Temperature gradient determination: When the rate of temperature change satisfies When an alert is triggered, among which The gradient threshold is (3℃ / min≤β≤8℃ / min); Composite judgment: When the current fluctuation rate Furthermore, if the temperature gradient exceeds the limit, it is judged as an emergency fault (15%≤γ≤20%).

[0026] Preferably, the hybrid inference engine comprises: Rule-based reasoning unit: Configure impeller imbalance rules: >60% and If the speed is >4.5 mm / s, the impeller is considered unbalanced. Case Reasoning Unit: The case database is retrieved based on cosine similarity, with the similarity formula as follows: ; in For real-time feature vectors, For example, the feature vector of the case; Deep learning unit: Uses CNN-LSTM model to process spectral sequences.

[0027] Preferably, the self-evolution management module includes: Dual-threshold anomaly detector: When data continuously exceeds k times the standard deviation for a duration greater than t minutes, the Isolation Forest algorithm is triggered; Rule generator: Automatically generates candidate rules for new fault modes obtained from clustering, using the format "IF condition set THEN fault type".

[0028] Preferably, the full-dimensional knowledge base is stored using knowledge graph technology and entity-relationship modeling is achieved through the Neo4j graph database. Entities include blower components, fault types, feature parameters, and solutions, while relationships include fault association, feature mapping, and solution matching.

[0029] like Figure 2 As shown, preferably, the online diagnostic method for wastewater treatment plant blowers based on a knowledge base includes the following steps: S1, the edge end collects vibration, temperature and current data in real time at a sampling rate of 10kHz; S2, threshold screening, performs edge-level anomaly detection: Calculate the effective value of vibration If satisfied Then mark it as an exception; Calculate the temperature gradient If satisfied Then mark it as an exception; When current fluctuation rate Furthermore, if the temperature gradient exceeds the limit, an emergency alarm will be triggered directly; S3 performs preprocessing on outlier data: filtering, noise reduction, and normalization; S4, Extract vibration features: kurtosis 1x frequency amplitude and its proportion ; S5, execute three-level progressive reasoning: Rule-based reasoning: Matches to a pre-defined rule base to diagnose common faults; Case-based reasoning: A solution for reusing a case database by retrieving cases using cosine similarity; Deep learning inference: Analyzing complex faults using CNN-LSTM models; S6, using the DS evidence theory to fuse the results of the three-level inference, calculates the fault confidence: ; in To have faith in the lower realm, Let m(B) be the upper confidence bound and the basic probability assignment. S7 generates a visual report containing diagnostic conclusions, supporting evidence, and maintenance recommendations.

[0030] Preferably, the rule reasoning in step S5 specifically includes: When both conditions are met >60% and When the speed is >4.5 mm / s, it is determined that the impeller is unbalanced; When temperature gradient >5℃ / min and current fluctuation rate When the value is greater than 15%, it is determined that the bearing is overheated.

[0031] Preferably, in the case reasoning of step S5, a similarity threshold is set. When the similarity between the real-time feature vector and the historical case is greater than the similarity threshold, the case solution is directly reused.

[0032] Preferably, in the deep learning inference of step S5: The CNN layer uses 3×3 convolutional kernels to extract vibration spectrum features; The LSTM layer uses 128 neurons to process time series and outputs a fault probability distribution.

[0033] Preferably, it also includes a knowledge self-evolution step: When the data continuously exceeds k times the standard deviation and lasts for more than t minutes, the Isolation Forest algorithm is triggered to detect outliers. Novel fault modes are discovered through hierarchical clustering, and candidate rules "IF condition set THEN fault type" are generated. After expert review, the knowledge base is updated and incremental training of the model is triggered.

[0034] Example 2: This embodiment provides an example of the application of the knowledge base-based online diagnostic system for wastewater treatment plant blowers of the present invention in a specific wastewater treatment plant. The process is as follows: I. System Deployment and Implementation: Hardware deployment: Accelerometers, thermistors, and current sensors are installed in key parts of the blower and connected to the network through a wireless edge gateway; Software configuration: Cloud-based knowledge graph database Neo4j, hybrid inference engine, and model training platform TensorFlow; User-side applications: Maintenance personnel can view diagnostic results through web and mobile app interfaces, and can also obtain repair guidance by combining AR glasses.

[0035] II. Example of Diagnostic Process: An accelerometer was installed at a wastewater treatment plant to monitor the vibration of a centrifugal blower. The waveform spectrum of a blower with an effective vibration value of 8.5 mm / s is shown below. Figure 3 As shown; the specific data diagnostic processing flow is as follows: Edge processing: Vibration characteristics were calculated. The 1x frequency amplitude accounted for 72%, and the kurtosis was 4.1, which was judged as abnormal data. Rule-based reasoning: Matching the rule "1x Rotation frequency ratio > 60% and vibration velocity > 4.5 mm / s → impeller imbalance", the preliminary diagnosis is impeller failure; Case Reasoning: Search the case database and find a historical case with 91% similarity: Imbalance caused by mud accumulation; Deep learning inference: The CNN-LSTM model outputs an impeller imbalance probability of 89%; Evidence fusion: After calculation using DS evidence theory, the confidence level for determining the fault as "impeller imbalance" is 93%. Report generation: Recommends shutdown and impeller cleaning, and displays rule matching details, historical case waveform comparison charts, and model heatmap.

[0036] III. Diagnostic Process Example: 1. Anomaly-driven knowledge discovery engine: A dual-threshold anomaly detection mechanism is constructed: when real-time data deviates from the normal range, such as when the effective value of vibration exceeds a preset standard by a preset multiple of the standard deviation and the duration exceeds a preset time, the knowledge discovery process is triggered. Outliers are identified through the isolated forest algorithm, and potential fault modes are mined by hierarchical clustering analysis. For example, a wastewater treatment plant discovered a new type of fault through this mechanism—the coating of the Roots blower rotor peeling off due to excessive intake humidity, and the knowledge base was automatically updated.

[0037] 2. Human-machine collaborative knowledge update closed loop: A complete process of "data monitoring - model training - expert review - knowledge storage" is established, and newly discovered fault modes automatically generate candidate rules; for example: IF humidity > 95% and high-frequency modulation signal appears during vibration, THEN rotor coating failure, which is pushed to the expert platform for verification.

[0038] 3. The specific process is as follows: Anomaly detection: The system detected vibration data with a high-frequency modulation signal exceeding the normal range for a fixed number of consecutive times, and the duration exceeded the preset time. Knowledge mining: Discovering new failure modes through cluster analysis; Rule generation: Automatically create candidate rules; Expert review: The application is pushed to the expert platform, where it is confirmed, the knowledge base is updated, and the model is retrained.

Claims

1. A knowledge-based online diagnostic system for wastewater treatment plant blowers, characterized in that, include: Edge-cloud collaborative architecture: consists of edge devices deployed at the device site and cloud servers; Cloud servers include: The knowledge base management system is used to build a full-dimensional knowledge base, including eight modules: storage device basic library, fault mode library, monitoring technology library, case experience library, maintenance strategy library, standard specification library, operating parameter library, and sensor integration library. It adopts knowledge graph structured storage. Hybrid reasoning engine: Performs a three-level progressive diagnosis on a cloud server, which combines rule-based reasoning, case-based reasoning, and deep learning reasoning, and fuses the results through DS evidence theory; The model training platform includes a self-evolution management module: optimizing the knowledge base and AI model through an anomaly-driven knowledge discovery mechanism and federated learning.

2. The online diagnostic system for wastewater treatment plant blowers based on a knowledge base according to claim 1, characterized in that, The edge end is configured with: Accelerometers, thermistors, and current sensors; Preprocessing module: Performs filtering, noise reduction, and normalization processing; Feature extraction module: Calculates the effective value of vibration. , cliff 1x frequency amplitude and its proportion ,in where n is the harmonic order; Threshold judgment module: This module compares the collected data with preset thresholds in the runtime parameter library in real time, performing initial anomaly screening. Specifically, it includes: Vibration threshold judgment: When the effective value of vibration meets the following conditions... When an alert is triggered, among which This is the vibration reference value under normal operating conditions of the equipment. For safety factor; Temperature gradient determination: When the rate of temperature change satisfies When an alert is triggered, among which Gradient threshold; Composite judgment: When the current fluctuation rate Furthermore, if the temperature gradient exceeds the limit, it is determined to be an emergency fault.

3. The online diagnostic system for wastewater treatment plant blowers based on a knowledge base as described in claim 1, characterized in that, The hybrid inference engine includes: Rule-based reasoning unit: Configure impeller imbalance rules: > 60% and If the speed is > 4.5 mm / s, the impeller is considered unbalanced. Case Reasoning Unit: The case database is retrieved based on cosine similarity, with the similarity formula as follows: ； in For real-time feature vectors, For example, the feature vector of the case; Deep learning unit: Uses CNN-LSTM model to process spectral sequences.

4. The online diagnostic system for wastewater treatment plant blowers based on a knowledge base as described in claim 1, characterized in that, The self-evolution management module includes: Dual-threshold anomaly detector: When data continuously exceeds k times the standard deviation for a duration greater than t minutes, the Isolation Forest algorithm is triggered; Rule generator: Automatically generates candidate rules for new fault modes obtained from clustering, using the format "IF condition set THEN fault type".

5. The online diagnostic system for wastewater treatment plant blowers based on a knowledge base according to claim 1, characterized in that, The full-dimensional knowledge base is stored using knowledge graph technology and uses the Neo4j graph database to implement entity-relationship modeling. Entities include blower components, fault types, feature parameters, and solutions, while relationships include fault association, feature mapping, and solution matching.

6. A knowledge-based online diagnostic method for wastewater treatment plant blowers according to any one of claims 1-5, characterized in that, Includes the following steps: S1, the edge end collects vibration, temperature and current data in real time at a sampling rate of 10kHz; S2, threshold screening, performs edge-level anomaly detection: Calculate the effective value of vibration If satisfied Then mark it as an anomaly; Calculate the temperature gradient If satisfied Then mark it as an anomaly; When current fluctuation rate Furthermore, if the temperature gradient exceeds the limit, an emergency alarm will be triggered directly; S3 performs preprocessing on outlier data: filtering, noise reduction, and normalization; S4, Extract vibration features: kurtosis 1x frequency amplitude and its proportion ; S5, execute three-level progressive reasoning: Rule-based reasoning: Matches to a pre-defined rule base to diagnose common faults; Case-based reasoning: A solution for reusing a case database by retrieving cases using cosine similarity; Deep learning inference: Analyzing complex faults using CNN-LSTM models; S6, using the DS evidence theory to fuse the results of the three-level inference, calculates the fault confidence: ； in To have faith in the lower realm, Let m(B) be the upper confidence bound and the basic probability assignment. S7 generates a visual report containing diagnostic conclusions, supporting evidence, and maintenance recommendations.

7. The online diagnostic method for wastewater treatment plant blowers based on a knowledge base according to claim 6, characterized in that, The rule reasoning in step S5 specifically includes: When both conditions are met > 60% and When the speed is > 4.5 mm / s, it is determined that the impeller is unbalanced; When temperature gradient >5℃ / min and current fluctuation rate If the temperature exceeds 15%, the bearing is considered overheated.

8. The online diagnostic method for wastewater treatment plant blowers based on a knowledge base according to claim 6, characterized in that, In the case reasoning of step S5, a similarity threshold is set. When the similarity between the real-time feature vector and the historical case is greater than the similarity threshold, the case solution is directly reused.

9. The online diagnostic method for wastewater treatment plant blowers based on a knowledge base according to claim 6, characterized in that, In the deep learning inference of step S5: The CNN layer uses 3×3 convolutional kernels to extract vibration spectrum features; The LSTM layer uses 128 neurons to process time series and outputs a fault probability distribution.

10. The online diagnostic method for wastewater treatment plant blowers based on a knowledge base according to claim 6, characterized in that, It also includes the steps of knowledge self-evolution: When the data continuously exceeds k times the standard deviation and lasts for more than t minutes, the Isolation Forest algorithm is triggered to detect outliers. Novel fault modes are discovered through hierarchical clustering, and candidate rules "IF condition set THEN fault type" are generated. After expert review, the knowledge base is updated and incremental training of the model is triggered.

Images