Multi-sensor fusion intelligent diagnosis method under multi-protocol communication architecture
By constructing a multi-protocol communication architecture and data fusion technology, the problem of sensor data silos has been solved, enabling unified processing and intelligent diagnosis of heterogeneous sensor data, thereby improving the accuracy of fault identification and the comprehensiveness of equipment status monitoring.
Patent Information
- Application Number
- CN202511202059.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-26
- Publication Date
- 2025-12-09
AI Technical Summary
Existing monitoring systems only support a single communication protocol, making it difficult to connect and integrate sensors from different manufacturers and of different types, resulting in data silos. There is a lack of effective fusion of multi-source sensor data, and single sensor data has limitations and uncertainties, leading to low fault identification accuracy.
A multi-protocol communication architecture is constructed, which accesses heterogeneous sensor data through multi-protocol communication modules. An adaptive protocol parsing algorithm is used to unify the data into a standard format, and wavelet denoising and normalization processing are performed. By combining BP neural network and DS evidence theory, the weight coefficients and warning thresholds are dynamically adjusted to achieve multi-sensor data fusion and intelligent diagnosis.
It has achieved effective aggregation and fusion of heterogeneous sensor data, improved the fault identification accuracy from 75% to over 95%, reduced the impact of false detection by a single sensor, and provided comprehensive equipment status monitoring and precise diagnosis.
Smart Images
Figure CN121094201A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application belongs to the technical field of intelligent monitoring and fault diagnosis of power equipment, and particularly relates to a multi-sensor fusion intelligent diagnosis method under a multi-protocol communication architecture. BACKGROUND
[0002] In recent years, with the deep implementation of industrial 4.0 and intelligent manufacturing concepts, industrial production is moving towards a highly automated and intelligent direction. Under this trend, the complexity of various industrial equipment has greatly increased, and a single sensor has been difficult to meet the requirements of all-around state monitoring of the equipment. For example, in order to ensure the safety and stability of the production process of chemical production equipment, not only the temperature and pressure of the key parts of the equipment need to be monitored, but also the material flow and composition information need to be obtained, so it is inevitable for multiple types of sensors to work together.
[0003] Most of the existing monitoring systems only support a single communication protocol. However, in actual applications, different manufacturers produce sensors and different types of sensors often use different communication protocols, such as RS485, CAN, Ethernet, ModbusRTU, etc. This results in great difficulties in connecting these heterogeneous sensors to the monitoring system. The data of different protocols cannot be effectively integrated and shared, forming a data island. Moreover, the traditional monitoring method mainly relies on the data of a single sensor for equipment state judgment, lacks effective fusion of multi-source sensor data, and the data of a single sensor can only reflect the state information of a certain aspect of the equipment, which has limitations and uncertainties.
[0004] Therefore, it is necessary to design a corresponding technical solution to solve the problem. SUMMARY
[0005] The technical problem to be solved by the present application is to provide a multi-sensor fusion intelligent diagnosis method under a multi-protocol communication architecture, to solve the data island problem mentioned in the background technology and improve the fault recognition accuracy.
[0006] To solve the above technical problems, the technical solution adopted by the present application is: A multi-sensor fusion intelligent diagnosis method under a multi-protocol communication architecture, comprising the following steps: S1, a multi-protocol communication architecture including a hardware layer and a protocol conversion layer is constructed, and the hardware layer includes a multi-protocol communication module. S2, access the heterogeneous sensor data of the power cabinet, demagnetization device and magnetic field circuit breaker in the pumped storage unit excitation system through the multi-protocol communication module, uniformly convert different protocol data into a standardized format through the adaptive protocol analysis algorithm of the protocol conversion layer, and then wavelet denoise the standardized data to filter out high-frequency noise; adopt normalization processing to eliminate dimensional differences; dynamically calculate weight coefficients according to sensor measurement errors, fuse multi-source data, and reduce the influence of single sensor measurement errors.
[0007] S3, extract time domain, frequency domain and trend characteristics, wherein the time domain characteristics include mean, variance and kurtosis, the frequency domain characteristics are extracted by FFT transformation of the signal main frequency component, and the trend characteristics are temperature change rate and current fluctuation coefficient; input the fused feature vector into the BP neural network trained by historical fault data, output the fault type probability value, and then convert it into a mass function to calculate the joint trust degree of multi-sensor diagnosis evidence.
[0008] S4, input the feature vector extracted in S3 into the intelligent diagnosis model combined by the BP neural network and the D-S evidence theory, realize the classification of the device state and the positioning of the fault, and push the early warning information through the dynamically adjusted multi-level early warning threshold.
[0009] S5, dynamically adjust the early warning threshold combined with the device operation history data. Preferably, the multi-protocol communication module supports RS485, CAN, Ethernet and Modbus RTU protocol, realizes electrical connection with the sensor through DB9 and RJ45 physical interfaces, and is used for accessing the heterogeneous sensor data of the power cabinet, demagnetization device and magnetic field circuit breaker in the pumped storage unit excitation system.
[0010] Preferably, the heterogeneous sensor includes the bridge arm CT, PT100 temperature sensor and thyristor patch temperature sensor of the power cabinet, the Hall element and demagnetization resistance temperature sensor of the demagnetization device, and the vibration sensor and temperature sensor of the magnetic field circuit breaker split coil.
[0011] Preferably, the adaptive protocol analysis algorithm supports Modbus RTU, CAN bus and Ethernet MMS to JSON format conversion.
[0012] Preferably, the heterogeneous sensor data collected by the multi-protocol communication module is subjected to wavelet denoising and normalization processing, and then the adaptive weighted fusion algorithm is used to calculate the weight and fuse multi-source data according to the formula is the weight coefficient of the jth sensor, is the weight coefficient of the jth sensor, is the measurement error of the jth sensor, is the number of sensors participating in fusion.
[0013] Preferably, the extracted time domain features include mean, variance and kurtosis, the frequency domain features are extracted by FFT transformation to extract the main frequency component of the signal, and the trend features are temperature change rate and current fluctuation coefficient.
[0014] Preferably, the intelligent diagnosis model comprises a BP neural network and a D-S evidence theory, wherein the BP neural network outputs probability values of fault types such as thyristor overheating, de-excitation resistor aging and mechanical damage of tripping coil after being trained by historical fault data, and the D-S evidence theory calculates a joint confidence level by fusing multiple sensor diagnosis results through a mass function. Preferably, in S5, the method for dynamically adjusting the early warning threshold in combination with the equipment operation history data is as follows: real-time correction is performed according to the equipment operation history data, when the joint confidence level is greater than 0.8, it is determined that a fault occurs, when the joint confidence level is between 0.5 and 0.8, it is determined that a warning occurs, and when the joint confidence level is less than 0.5, it is determined that the equipment is normal;The equipment operation history data includes, for example, unit start-stop cycle and load change curve.
[0015] Preferably, the early warning information is pushed in real time through a short message and an in-station monitoring system, and an operation and maintenance work order containing the fault type, location and recommended measures is automatically generated.
[0016] A multi-sensor fusion intelligent diagnosis system under a multi-protocol communication architecture adopts the multi-sensor fusion intelligent diagnosis method under the multi-protocol communication architecture.
[0017] The present application can achieve the following beneficial effects: 1. The multi-sensor fusion intelligent diagnosis method under the multi-protocol communication architecture can effectively access heterogeneous sensor data on devices such as power cabinets, de-excitation devices and magnetic field circuit breakers through the constructed multi-protocol communication architecture, integrated communication modules supporting multiple protocols such as RS485, CAN, Ethernet and Modbus RTU, and convert them into a standardized data format through a self-adaptive protocol analysis algorithm, which enables originally scattered and isolated data to be converged and fused, providing a comprehensive and accurate data basis for subsequent comprehensive analysis and diagnosis, and fundamentally solving the data island problem. 2. The multi-sensor fusion intelligent diagnosis method under the multi-protocol communication architecture can dynamically adjust the weight according to the sensor measurement error through a self-adaptive weighted fusion algorithm by adopting a multi-sensor fusion technology, fuse the data of multiple sensors, and at the same time, combine multiple feature extraction methods and intelligent diagnosis models to comprehensively analyze and judge the fault, which greatly reduces the influence of single sensor measurement error on the diagnosis result and effectively improves the fault recognition accuracy, from 75% of single sensor to more than 95%. BRIEF DESCRIPTION OF DRAWINGS
[0018] The present application will be further described below in combination with the drawings and embodiments: Fig. 1 This is a diagram of the multi-protocol communication architecture of the present invention; Fig. 2 This is a flowchart of the multi-sensor fusion algorithm of the present invention; Fig. 3 This is a diagram of the intelligent diagnostic model architecture of the present invention. Detailed Implementation
[0019] Preferred solutions include Figs. 1 to 3 As shown, a multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture is described. The specific method is as follows: A multi-protocol communication architecture comprising a hardware layer and a protocol conversion layer is constructed. The hardware layer integrates multi-protocol communication modules supporting RS485, CAN, Ethernet, and Modbus RTU protocols to access heterogeneous sensor data from the power cabinet, demagnetizing device, and field circuit breaker in the pumped storage unit's excitation system. The protocol conversion layer uses an adaptive protocol parsing algorithm to unify different protocol data into a standardized format. Wavelet denoising and normalization preprocessing are applied to the standardized data to eliminate dimensional differences and noise interference. An adaptive weighted fusion algorithm is then used to fuse the preprocessed data, dynamically calculating weighting coefficients based on sensor measurement errors. The formula is as follows: Calculate weights and fuse multi-source data, where For the first The weighting coefficients of each sensor Let j be the measurement error of the j-th sensor. The number of sensors participating in the data fusion is specified. Time-domain, frequency-domain, and trend features are extracted from the fused data and input into an intelligent diagnostic model combining a BP neural network and DS evidence theory to achieve equipment status classification and fault location. Furthermore, warning information is pushed through dynamically adjusted multi-level warning thresholds.
[0020] Furthermore, the heterogeneous sensors include bridge arm CTs, PT100 temperature sensors, SCR patch temperature sensors, Hall elements and transmitters of demagnetizing devices, and vibration and temperature sensors of the trip coils of magnetic field circuit breakers. The adaptive protocol parsing algorithm supports the conversion of Modbus RTU, CAN bus, Ethernet MMS to JSON format.
[0021] Further, the time domain features include mean, variance, kurtosis, the frequency domain features are extracted by FFT transformation to extract the signal main frequency component, the trend features include temperature change rate and current fluctuation coefficient, the BP neural network inputs the fused feature vector and outputs the fault type probability value, and the D-S evidence theory fuses the mass function to fuse the multi-sensor diagnosis results to calculate the joint trust degree. For example, when the current variance of the de-excitation resistor branch exceeds the threshold value, the current sharing abnormality is prompted, and the time domain signal is converted into the frequency domain through the FFT transformation to extract the main frequency component. If the main frequency of the tripping coil vibration signal doubles, the mechanical resonance risk is prompted. Meanwhile, the temperature change rate (unit: ℃ / min) and the current fluctuation coefficient (calculation formula: standard deviation / mean) are used to identify the abnormal parameter change trend. For example, when the temperature of the thyristor rises by more than 20 ℃ within 10 minutes, the early warning is triggered.
[0022] Further, the multi-protocol communication module is used to access the RS485, CAN, Ethernet and Modbus RTU protocol sensors to collect the sensor data of the power cabinet, de-excitation device and magnetic field circuit breaker. The data preprocessing module is used to perform wavelet denoising and normalization processing on the collected data. The adaptive weighted fusion module is used to calculate the weight and fuse the multi-source data according to the formula: wherein w i is the weight coefficient of the i th sensor, e j is the measurement error of the j th sensor, and n is the number of sensors participating in fusion. The feature extraction module is used to extract the time domain, frequency domain and trend features. The intelligent diagnosis module includes the BP neural network and the D-S evidence theory, which is used to realize the state classification, fault positioning and dynamic early warning according to the features. The BP neural network training process of the BP neural network and the D-S evidence theory fusion diagnosis is as follows: the input layer receives the fused feature vector such as [current mean, temperature change rate, current sharing degree]; the hidden layer adopts the ReLU activation function, and the network is trained by the historical fault data such as the thyristor overheating, de-excitation resistor aging and tripping coil cracking sample, and the probability value of each fault type is output.
[0023] Further, the BP neural network is trained by the historical fault data to output the probability value of the fault types such as the thyristor overheating, de-excitation resistor aging and tripping coil mechanical damage. The D-S evidence theory fuses the joint trust degree by fusing the multi-sensor evidence to improve the diagnosis accuracy. The fault probability output by the BP neural network is converted into the mass function, and the diagnosis evidences of multiple sensors such as the current abnormality evidence and the temperature abnormality evidence are fused. For example, if the mass value of the current sensor diagnosing “overheating” is 0.7 and the mass value of the temperature sensor diagnosing “overheating” is 0.6, the joint trust degree is calculated by the Dempster combination rule.
[0024] Working principle: The multi-sensor fusion intelligent diagnosis method under the multi-protocol communication architecture takes the excitation system of pumped storage unit as an example. In the excitation system of pumped storage unit, various types of sensors are deployed in the power cabinet, de-excitation device and magnetic field circuit breaker opening coil. The power cabinet includes bridge arm CT for current monitoring, ModbusRTU protocol; PT100 temperature sensor for temperature monitoring, RS485 protocol; thyristor patch temperature sensor, RS485 protocol. The de-excitation device includes Hall element for excitation current monitoring, 4-20mA analog, which needs to be converted to ModbusRTU or Ethernet protocol by transducer; de-excitation resistance temperature sensor, RS485 protocol. The magnetic field circuit breaker includes opening coil vibration sensor, CAN bus protocol; temperature sensor, RS485 protocol.
[0025] The hardware layer integrates the multi-protocol communication module of RS485, CAN, Ethernet and ModbusRTU, which directly accesses the above sensors through physical interfaces such as DB9 and RJ45, realizing parallel collection of heterogeneous data.
[0026] The protocol conversion layer standardizes the original data of the sensor. The original data carries different communication protocols, such as ModbusRTU 16-bit data, CAN bus 29-bit identifier data and Ethernet MMS TCP / IP data packet. The protocol conversion layer realizes automatic detection of data frame header through adaptive protocol analysis algorithm, matches the pre-defined protocol analysis rules, uniformly converts different protocol data into JSON format, and outputs standardized data containing device ID, sensor type, measurement value and timestamp. Through this process, the protocol barrier is eliminated, and a unified data interaction language is formed.
[0027] Then the data is preprocessed, including denoising and normalization. For high-frequency noise in current, temperature and other signals, wavelet transform is used to decompose the signal into different frequency layers, retain the low-frequency effective signal and filter out the high-frequency noise. At the same time, different dimension data is converted to [0, 1] interval, the formula is: ; Among them, and are the historical measurement extreme values of the sensor, which ensure the numerical stability of the subsequent fusion calculation.
[0028] The adaptive weighted fusion algorithm dynamically allocates weights according to the historical measurement error of the sensor, which is obtained by long-term monitoring and statistics, and gives priority to high-precision sensor data. The weight calculation formula is: Calculate the weight and fuse multi-source data, where is the Weight coefficient of each sensor, Measurement error of the jth sensor, Number of sensors participating in fusion. Through this algorithm, multi-source sensor data is fused into comprehensive feature values, reducing the influence of single sensor measurement errors.
[0029] Multi-dimensional feature extraction extracts three types of features from the fused data: time domain features, including mean, reflecting the average level of the signal; variance, reflecting the degree of fluctuation; kurtosis, reflecting the impact characteristics of the signal. For example, when the current variance of the de-excitation resistor branch exceeds the threshold, it indicates an abnormal current sharing. Frequency domain features are extracted by converting time domain signals to frequency domain through FFT transformation, such as the main frequency component. If the main frequency of the breaking coil vibration signal doubles, it indicates a risk of mechanical resonance. Trend features include temperature change rate (unit: ℃ / min) and current fluctuation coefficient (formula: standard deviation / mean), which are used to identify abnormal parameter change trends, such as triggering a warning when the thyristor temperature rises by more than 20℃ in 10 minutes.
[0030] In the BP neural network and D-S evidence theory fusion diagnosis, the input layer of the BP neural network training is the fused feature vector, such as [current mean, temperature change rate, current sharing degree]. The hidden layer uses the ReLU activation function, and the network is trained with historical fault data such as thyristor overheating, de-excitation resistor aging, and breaking coil cracking samples, outputting probability values for each fault type. The fault probability output by the BP neural network is converted into a mass function, and the diagnostic evidence of multiple sensors is fused, such as current abnormal evidence and temperature abnormal evidence. For example, if the mass value of the current sensor diagnosing "overheating" is 0.7 and the mass value of the temperature sensor diagnosing "overheating" is 0.6, the joint confidence is calculated by the Dempster combination rule: ; When the joint confidence is greater than 0.8, it is determined as a fault, 0.5-0.8 as a warning, and less than 0.5 as normal, avoiding misjudgment of a single sensor or algorithm.
[0031] The dynamic warning mechanism combines with the historical data of the equipment, such as the unit start-stop cycle and load change curve, to dynamically adjust the warning threshold. For example, when the unit is running normally, the thyristor temperature warning threshold is set to 90℃; during the unit start-stop process, due to the large current fluctuation, the threshold is temporarily adjusted to 95℃, reducing false positives. Warning information is pushed in real time through SMS and station monitoring system, and an operation and maintenance work order is generated.
[0032] Through the above principles, the system realizes full-state perception, multi-dimensional analysis and accurate diagnosis of key equipment of the pumped storage unit excitation system, improves the fault identification accuracy from 75% of a single sensor to more than 95%, and provides core technical support for intelligent operation and maintenance of power stations.
[0033] The above embodiments are only preferred technical solutions of the present application, and should not be regarded as a limitation of the present application. The protection scope of the present application should be the technical solutions recited in the claims, including equivalent replacement solutions of the technical features recited in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present application.
Claims
1. A multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture, characterized in that... Includes the following steps: S1. Construct a multi-protocol communication architecture that includes a hardware layer and a protocol conversion layer. The hardware layer includes a multi-protocol communication module. S2. Access heterogeneous sensor data from power cabinets, demagnetizing devices, and field circuit breakers in the pumped storage unit excitation system via a multi-protocol communication module. Then, use the adaptive protocol parsing algorithm of the protocol conversion layer to unify the different protocol data into a standardized format. Finally, perform wavelet noise reduction on the standardized data to filter out high-frequency noise. Normalization is used to eliminate dimensional differences; weighting coefficients are dynamically calculated based on sensor measurement errors, and multi-source data are fused to reduce the impact of mismeasurement from a single sensor. S3. Extract time-domain, frequency-domain, and trend features. The time-domain features include mean, variance, and kurtosis. The frequency-domain features are extracted by FFT transformation to extract the main frequency component of the signal. The trend features are the temperature change rate and current fluctuation coefficient. The fused feature vector is input into a BP neural network trained with historical fault data to output the fault type probability value. Then, it is converted into a mass function and the joint confidence level is calculated by fusing multi-sensor diagnostic evidence. S4. Input the feature vector extracted in S3 into the intelligent diagnostic model that combines BP neural network and DS evidence theory to realize equipment status classification and fault location, and push early warning information through dynamically adjusted multi-level early warning thresholds; S5. Dynamically adjust the early warning threshold based on historical equipment operation data.
2. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 1, characterized in that: The multi-protocol communication module supports RS485, CAN, Ethernet, and Modbus RTU protocols. It achieves electrical connection with sensors through DB9 and RJ45 physical interfaces and is used to access heterogeneous sensor data from power cabinets, demagnetizing devices, and field circuit breakers in the excitation system of pumped storage units.
3. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 2, characterized in that: The heterogeneous sensors include the bridge arm CT of the power cabinet, the PT100 temperature sensor, the silicon controlled rectifier patch temperature sensor, the Hall element of the demagnetizing device, the demagnetizing resistor temperature sensor, and the vibration sensor and temperature sensor of the magnetic field circuit breaker trip coil.
4. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 1, characterized in that: The adaptive protocol parsing algorithm supports the conversion of Modbus RTU, CAN bus, Ethernet MMS to JSON format.
5. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 1, characterized in that: The heterogeneous sensor data acquired by the multi-protocol communication module is subjected to wavelet denoising and normalization, and then fused using an adaptive weighted fusion algorithm according to the formula. Calculate weights and fuse multi-source data, where For the first The weighting coefficients of each sensor Let j be the measurement error of the j-th sensor. The number of sensors participating in the fusion.
6. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 5, characterized in that: The extracted time-domain features include mean, variance, and kurtosis. The frequency-domain features are extracted by FFT transformation to extract the main frequency component of the signal. The trend features are the temperature change rate and the current fluctuation coefficient.
7. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 1, characterized in that: The intelligent diagnostic model includes a BP neural network and a DS evidence theory. The BP neural network is trained with historical fault data and outputs probability values for fault types such as thyristor overheating, magnetizing resistor aging, and mechanical damage to the trip coil. The DS evidence theory calculates the joint confidence level by fusing multi-sensor diagnostic results through a mass function.
8. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 1, characterized in that: In S5, the method for dynamically adjusting the early warning threshold based on historical equipment operation data is as follows: real-time correction is made based on historical equipment operation data. When the joint trust level is >0.8, it is judged as a fault; 0.5-0.8 is an early warning; and <0.5 is normal. Historical equipment operation data includes unit start-up and shutdown cycles and load change curves.
9. The multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture according to claim 8, characterized in that: Early warning information is pushed in real time via SMS and the on-site monitoring system, and maintenance work orders containing fault type, location and suggested measures are automatically generated.
10. A multi-sensor fusion intelligent diagnostic system under a multi-protocol communication architecture, characterized in that: The method employs a multi-sensor fusion intelligent diagnostic method under a multi-protocol communication architecture as described in any one of claims 1-9.