Sensor fault diagnosis method

Abstract

The invention discloses a sensor fault diagnosis method, which comprises the following steps of: reading acquired data of a sensor in real time, and recording the data as a sensor reading at the current moment; judging whether the obtained sensor reading is within a preset threshold range or not; calculating sensor reading increments of adjacent time points, constructing increment distribution histograms of a plurality of time points, counting frequencies of different increment intervals, extracting increment distribution characteristic values, judging a ratio relation of the increment distribution characteristic values, and performing moving average processing and regression analysis on all sensor readings to obtain a plurality of sensor reading values; calculating a regression coefficient R to evaluate the linear relation strength between the data; comparing the Rvalue with a preset regression coefficient value; and judging whether the regression coefficient R is smaller than the preset value. Accurate diagnosis and positioning of sensor faults are realized, the false alarm rate is reduced, the reliability of a detection result is improved, and the method has universality, can be suitable for various types of environmental sensors, and meets different environmental monitoring requirements.

Description

Technical Field

[0001] The present invention belongs to the technical field of environmental monitoring sensor detection, and in particular relates to a sensor fault diagnosis method. Background Art

[0002] Existing sensor fault diagnosis faces many challenges. First, most sensor detection relies on a single threshold to judge and identify sensor faults, which is easily affected by environmental noise and data fluctuations, resulting in false positives and false negatives. Second, the lack of a multi-level diagnostic mechanism makes it difficult to timely identify and locate complex faults (such as abnormal sensor position, occlusion, and communication failure). Third, traditional methods are usually unable to monitor the dynamic operating status of sensors in real time, making it difficult to detect and eliminate potential faults in a timely manner, reducing their reliability and data accuracy. In addition, existing methods have poor versatility and cannot adapt to the fault diagnosis needs of sensors in different environments. In short, existing methods are difficult to ensure the stable operation of sensors in complex and dynamic environments, which in turn affects the accuracy and effectiveness of monitoring results. Summary of the invention

[0003] In order to solve the technical problem mentioned in the background technology, the present invention provides a sensor fault diagnosis method.

[0004] The present invention adopts the following technical solution.

[0005] A sensor fault diagnosis method, comprising the steps of: Step S1: Acquire the data of the target environment sensor, define the normal working range of the sensor, including the threshold range and data change characteristics; read the collected data T of the sensor in real time, and record the data as the sensor reading at the current moment; Step S2: Determine whether the obtained sensor reading is within a preset threshold range; when the sensor reading is not within the preset range, output an abnormal warning, indicating that there may be a sensor communication failure or a sensor hardware failure; otherwise, continue to analyze the change trend of the data; Step S3: Calculate the sensor reading increments ΔT at adjacent time points, construct the increment distribution histogram of multiple time points, count the frequencies of different increment intervals, and extract the increment distribution feature values, such as the 95% and 100% increment thresholds (ΔX 95 and ΔX 100 ); Step S4: Determine the ratio relationship of the incremental distribution characteristic values, i.e. ΔX 95 / ΔX 100 Whether the preset conditions are met; if the ratio is less than the preset value, an abnormal warning is output, indicating that the sensor may have position abnormality, occlusion or other external interference problems; otherwise, continue to perform subsequent fault diagnosis steps; Step S5: Perform sliding average processing on all sensor readings to eliminate obviously abnormal fault data; Step S6: Perform regression analysis on the smoothed data and calculate the regression coefficient R 2 To evaluate the strength of the linear relationship between the data; compare the R2 value with the preset regression coefficient preset value; determine the regression coefficient R 2 Is it less than the preset value? If it is less than the preset value, an abnormal warning is output to indicate that the sensor may be faulty, resulting in abnormal data fluctuations; otherwise, the sensor is output to indicate that it is operating normally.

[0006] The present invention constructs a multi-level fault diagnosis mechanism, gradually and deeply analyzes sensor data, from preliminary range verification to complex incremental analysis and regression analysis, to achieve timely identification and location of complex faults. First, the obvious abnormalities of the sensor, such as communication failures or hardware failures, are quickly detected through threshold range judgment; then, incremental analysis is used to construct an incremental distribution histogram, extract incremental eigenvalues, and identify abnormalities caused by external interference or sensor state changes, such as abnormal sensor position or occlusion; finally, abnormal data is eliminated through sliding average processing, and regression analysis is performed to deeply evaluate the linear relationship between data, further identify data fluctuation abnormalities, and locate potential faults. This multi-level diagnostic mechanism solves the problem that existing methods are difficult to identify and locate complex faults in a timely manner, and through layered analysis, it can accurately diagnose different fault types of sensors and reduce the probability of false alarms and missed alarms.

[0007] Preferably, the threshold range and incremental characteristic value used in the method can be adjusted according to the characteristics of different environmental sensors to meet the diagnostic requirements of various types of sensors, such as temperature, humidity, pressure, light intensity, air quality, etc.

[0008] Furthermore, the present invention achieves the versatility of multi-type sensor fault diagnosis by flexibly adjusting diagnostic parameters such as threshold ranges and incremental characteristic values ​​to adapt to different types of environmental sensors. In a specific implementation, the threshold range and incremental characteristic values ​​can be set according to the characteristics of the sensor type, so that the diagnostic mechanism can adapt to the working characteristics of different sensors. For example, refer to the technical specifications of the target environmental sensor and the parameters provided by the manufacturer. The manufacturer usually provides the normal working range of the sensor, including the minimum and maximum measurement values, operating temperature range, accuracy and response time, etc. These parameters can be used as a reference for the initial threshold range; the normal range of parameters such as temperature, humidity, and air quality in environmental monitoring may already have relevant standards or recommended values, and these standards can provide a reference for threshold setting; the threshold can also be determined in combination with the environmental characteristics of the sensor and the specific application scenario.

[0009] Preferably, the incremental distribution histogram in the present invention can find the incremental distribution characteristic values ​​from multiple angles, including the median, mode and mean, which are respectively used as the center points of the data distribution, and then the 95% ratio width calculation is performed. Specifically, first, the median of the sensor incremental data is calculated as the center point of the histogram; then, it is expanded to both sides from the center point, and the frequency of the incremental data is accumulated until the cumulative frequency reaches 95% of the entire data set, and the incremental value range corresponding to both sides is determined to obtain the 95% width centered on the median. Similarly, the 95% ratio width calculation is performed with the mode and mean as the center. For any of the above methods, if the calculated 95% width range exceeds the preset normal range, it can be determined that the sensor is faulty. Through this multi-center point determination method, the accuracy of fault diagnosis is further improved.

[0010] In addition, the present invention provides a variety of data smoothing methods such as sliding average processing, weighted sliding average, exponential smoothing, etc., which can be flexibly selected according to the characteristics of different sensors to improve the accuracy of abnormal data identification. Through this configurable, multi-level diagnosis mechanism, it can effectively respond to the fault diagnosis needs of different environmental sensors and improve the wide applicability of diagnosis.

[0011] Preferably, the present invention supports dynamic adjustment of data acquisition frequency and analysis parameters during the diagnosis process to adapt to different environments and changes in sensor status. The diagnosis results can be presented through a visual interface, providing intuitive sensor operating status and fault reports, so that users can manage and maintain according to the characteristics of different types of sensors.

[0012] Furthermore, the method of the present invention can be integrated with an existing environmental monitoring system to achieve unified management and fault diagnosis of various environmental sensors, providing reliable technical support for environmental monitoring and equipment management.

[0013] Furthermore, the present invention can also be applied in an Internet of Things platform to achieve remote monitoring of sensor status and fault alarms through network connections, and support centralized management of distributed sensor networks.

[0014] The present invention constructs a multi-level fault diagnosis mechanism, which analyzes sensor readings in depth layer by layer, from preliminary range verification to complex incremental analysis and regression analysis, gradually narrowing the scope of the fault.

[0015] Through sliding average processing and distribution histogram analysis, the present invention can remove noise and outliers in sensor readings, thereby effectively reducing the false alarm rate. Compared with simple threshold judgment methods, the diagnostic method of the present invention is more robust and can cope with fluctuations and instabilities in sensor readings in practical applications.

[0016] Beneficial effects: The entire solution adopts a multi-level diagnostic mechanism, from preliminary range verification to complex incremental analysis and regression analysis, to gradually and deeply identify sensor faults. This progressive diagnostic method helps to distinguish different types of faults; compared with the traditional single threshold judgment method, the present invention can more accurately identify and locate complex fault types; by introducing sliding average processing and incremental distribution histogram analysis, the present invention can effectively eliminate noise and outliers in the data, thereby reducing the false alarm rate. Dealing with fluctuations and noise in sensor data has always been a difficult point in monitoring. The present invention uses data smoothing and incremental analysis methods to dynamically adapt to data changes in practical applications and avoid excessive sensitivity to sudden abnormalities or noise signals.

[0017] The key problems solved by the present invention include how to accurately diagnose the faults of environmental sensors in complex and dynamic environments. Through a multi-level diagnostic mechanism, data smoothing processing and incremental feature analysis, accurate diagnosis and positioning of sensor faults are achieved, the false alarm rate is reduced, and the reliability of detection results is improved. At the same time, it has versatility and can be applied to various types of environmental sensors to meet the needs of different environmental monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the diagnostic method in the embodiment; Figure 2 is a histogram in the embodiment; Figure 3 This is a data graph after smoothing in the embodiment. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The temperature sensor is a representative environmental sensor, and this embodiment is described by taking the fault diagnosis of the temperature sensor as an example.

[0021] A temperature sensor fault diagnosis method is used to accurately and quickly diagnose the fault of an environmental temperature sensor. This method uses a multi-step diagnosis mechanism to effectively determine the fault type of the sensor and give corresponding processing suggestions. Figures 1 to 3 As shown, the following steps are included: Step S1: Get the temperature sensor reading, get the temperature data T of the current temperature sensor through the sensor interface, and record the data as the sensor reading at the current moment; For step S1: First, make sure that the temperature sensor has been correctly connected to the sensor interface module of the computer system (the computer-controlled environmental monitoring system) to ensure that the communication between the sensor and the data acquisition module is unimpeded. Once the connection is completed, the system will initialize the sensor and prepare it to enter the working state; Step S2: Define the temperature threshold T according to the preset environment or equipment requirements min and T max , then, check the current temperature data T to confirm whether it is within the valid range; Step S3: Calculate the temperature difference ΔT between the current time point and the previous time point, ΔT = T t+1 -T t , and accumulate the ΔT values ​​at multiple time points to construct incremental statistics, and use these data to generate a distribution histogram of temperature increments; Step S4: Based on the distribution histogram of the temperature increment, identify and extract the 95% and 100% increment thresholds (ΔX 95 and ΔX 100 ), then the ratio calculation (ΔX 95 / ΔX 100 ), and compare it with the preset threshold (0.8); Step S5: Perform sliding average processing on the sensor readings to eliminate obviously abnormal fault data; Step S6: Perform regression analysis on the smoothed data and calculate the regression coefficient R 2 To evaluate the strength of the linear relationship between the data, the R2 value was compared with the preset threshold (0.75).

[0022] In this embodiment, a data acquisition command is sent to the temperature sensor through a preset communication protocol (such as UART, 1-wire, etc.). The control sensor starts measuring the temperature of the current environment and prepares the measurement results for reading. After the sensor completes the temperature measurement, the temperature data is read from the sensor according to the communication protocol. These data are usually expressed in digital form, such as Celsius (℃) or Fahrenheit (°F), depending on the configuration and requirements of the sensor.

[0023] For step S2, according to the preset environment or equipment specifications, the effective range of temperature data is established, that is, the minimum threshold T is defined. min and the highest threshold T max, the threshold can be adjusted according to the environmental characteristics of the sensor and the specific application scenario. For example, the normal range of the temperature sensor in indoor and outdoor environments may be different, and it will be affected by the process flow in an industrial production environment. For different application scenarios, it is necessary to set the threshold range and data change characteristics that adapt to the scene characteristics in combination with the actual environmental conditions. Subsequently, the currently acquired temperature data T is automatically checked to confirm whether it falls within this preset valid range.

[0024] If the temperature data T does not meet T min ≤T≤T max If the temperature data T meets the validity check condition, that is, it is within the valid range, the data is automatically marked as abnormal, and a prompt message "Need to check sensor communication or hardware failure" is generated and output, and the diagnosis process is interrupted. If the temperature data T meets the validity check condition, that is, it is within the valid range, the data is accepted as valid input, and the subsequent data processing or analysis process continues without further intervention.

[0025] For step S3, a series of objective and quantitative operations are performed to analyze the changing trend of temperature data. The specific process is as follows: First, get the temperature reading T at the current time point (t+1) t+1 and the temperature reading T at the previous time point (t) t Then, based on these two data points, the temperature difference ΔT between them is calculated, that is, ΔT: ΔT = T t+1 -T t ,This step is the key to quantifying the temperature trend, which reflects the amount of temperature change between two consecutive time points.

[0026] In order to more fully understand the statistical characteristics of temperature changes, it is not enough to calculate the ΔT value at a single time point, but it is also necessary to accumulate the ΔT values ​​at multiple consecutive time points to form a series of temperature increment data. These data can reflect the specific situation of temperature fluctuations over a period of time. By performing statistical analysis on the accumulated ΔT values, various statistics such as mean, standard deviation, median, etc. can be calculated to quantify the overall trend and variability of temperature changes.

[0027] Generate a distribution histogram based on the accumulated ΔT values, and divide all the temperature increment data into certain intervals. Set the number of intervals in the increment range and the width of each interval. Usually the number of intervals can be based on the Sturgess formula or freely selected to ensure that the histogram has sufficient resolution and statistical significance. In the histogram, the horizontal axis represents the different intervals of temperature increment (that is, the different value ranges of ΔT), and the vertical axis represents the number or frequency of occurrences of ΔT values ​​falling in each interval. In this way, the central tendency, dispersion degree, and possible outliers or extreme cases of temperature increment can be clearly displayed. In order to more accurately identify sensor anomalies, histogram analysis is not limited to simple mean and variance analysis, but to find incremental distribution characteristic values ​​from multiple angles, including but not limited to median, mode, mean, etc. The specific implementation method is as follows: Centered on the median: First, calculate the median of the temperature increment data, that is, sort all the incremental data from small to large and take the data value in the middle. The median is a robust estimate of the position characteristics of the data set and can effectively resist the influence of outliers; centering on the median, expand to both sides to find the range that contains 95% of the incremental data, that is, find the upper and lower limits of the 95% range of the incremental data near the median. Count the data that falls outside this range. If the data point exceeds this range, it is considered that there may be anomalies; Centered on the mode: After constructing the histogram, find the interval with the highest frequency, and its corresponding incremental data is called the mode. The mode reflects the most common incremental characteristics of the data set. Centered on the mode, find the range that contains 95% of the incremental data, and determine the upper and lower limits of the 95% range near the mode. Through histogram analysis, if a large number of data points are outside this range, it indicates that there may be problems such as abnormal sensor position or occlusion; Centered on the mean: Calculate the mean of the incremental data, which is an overall reflection of the temperature change trend. The mean provides a global change feature while including all incremental data. Centered on the mean, calculate the width range that includes 95% of the incremental data. According to statistical principles, the upper and lower limits of the 95% interval can be determined by the standard deviation, that is, [Mean-1.96×σ, Mean+1.96×σ]. Check if the data point exceeds this range to determine whether there is an anomaly; Through the above multi-angle analysis, the 95% ratio width range centered on the median, mode, mean, etc. is obtained.

[0028] For step S4, firstly, the 95% and 100% increment thresholds (denoted as ΔX) are automatically determined based on the distribution histogram of the temperature increment. 95 and ΔX 100 ). Here, ΔX 95 Indicates that in the histogram, 95% of the ΔT values ​​are less than or equal to the threshold; and ΔX 100Then, it is a theoretical maximum value, which can be understood as the maximum ΔT value in the data set or a number close to it. Next, calculate the ratio of these two thresholds, that is, ΔX 95 / ΔX 100 . This ratio reflects one aspect of the concentration or variability of the temperature increment distribution. If the ratio is close to 1, it means that the distribution of temperature increments is relatively concentrated; if the ratio is small, it indicates that there is a large variability. The calculated ratio is compared with a preset threshold (0.8 in this embodiment). This preset threshold is set based on system requirements, environmental characteristics, historical data analysis or experience, and is used to evaluate the rationality or acceptability of the current temperature increment distribution; In order to further judge the status of the sensor, it is necessary to perform centrality analysis on the temperature increment distribution from multiple angles. First, judge based on the median: Based on the 95% ratio range centered on the median in the previous step, calculate the ratio ΔX centered on the median 95 / ΔX 100 . If this ratio is less than 0.8, it means that the distribution of temperature increments is not concentrated enough near the median, and there may be an abnormality. Next, judge based on the mode: If the ratio is less than 0.8, it means that the distribution of incremental data near the mode is abnormal, indicating that there may be improper sensor location or occlusion problems. Finally, judge based on the mean: If the ratio is less than 0.8, it means that the distribution of incremental data near the mean is not concentrated enough, and there may be a fault; Comprehensively judge, if any of the ratio judgments centered on the median, mode, and mean does not meet the preset conditions (that is, the ratio is less than 0.8), it is determined that the distribution characteristics of the current temperature increment may be abnormal. This means that there may be a sensor failure, such as improper sensor position, occlusion, or interference from other external factors. Once any ratio does not meet the conditions, the prompt message "Need to check the sensor position or occlusion" is immediately output, and the current diagnostic process is stopped, marking that the temperature sensor is not working properly. If all ratios meet the preset conditions (that is, they are all greater than or equal to 0.8), it is considered that the distribution of the current temperature increment is reasonable or acceptable under multi-angle analysis, and the subsequent diagnostic steps are continued.

[0029] For step S5, the raw sensor readings are smoothed by applying a sliding average method (combined with a sliding window technique) to improve the quality and stability of the data. First, you need to determine the size of the sliding window, that is, the number of consecutive data points considered when smoothing. The choice of window size depends on the characteristics of the data and the goal of the smoothing process. Larger windows can smooth out outliers and noise more effectively, but may cause the data series to respond more slowly and mask some important details. Smaller windows can retain more details, but the smoothing effect on outliers may be weaker. The sliding average method replaces the value of the center point of the window by calculating the average of all data points in the sliding window. As the window moves over the data sequence, each data point is replaced by the average of the data points around it, thereby achieving data smoothing. Specifically, for each window position, the arithmetic mean of all data points in the window is calculated and assigned to the data point in the center of the window.

[0030] The sliding average method itself does not directly identify or remove outliers, but it indirectly reduces the impact of outliers on the overall data by calculating the average. In the smoothed data series, outliers are usually "diluted" by the average of their neighboring normal values, thereby reducing their weight in data analysis.

[0031] After smoothing is complete, the smoothing effect can be evaluated by comparing the differences between the original data and the smoothed data. If the smoothing process effectively reduces the noise and outliers in the data while retaining the main trends and characteristics of the data, then the smoothing process can be considered successful.

[0032] For step S6, regression analysis is performed on the temperature data after sliding average processing to determine the strength of the linear relationship between the data and ultimately determine the operating status of the temperature sensor. The specific process of this step includes data preparation, linear regression analysis, regression coefficient calculation, and fault judgment. The implementation process of this step is described in detail below: In step S5, the temperature data has been processed by sliding average to obtain a set of smoothed temperature data series, which reflects the changing trend of temperature over a period of time, and has eliminated outliers and noise, and has good smoothness and stability; The smoothed temperature data series is constructed as a time series data set, and each point in the data set is set to (t i ,T i ), where t i represents the time point, T i Represents the smoothed temperature value. This time series will be used as the input data for regression analysis to evaluate the relationship between temperature and time. Assume that there is a certain linear relationship between temperature data and time, that is, temperature T can be expressed as a linear function of time t: T = a·t+b; where a is the slope, representing the rate of change of temperature over time; b is the intercept, representing the temperature value when time is zero. The data is linearly fitted by the least squares method to find the best linear equation to minimize the error between the temperature data points and the fitted straight line. The goal of the least squares method is to solve the parameters a and b so that the following objective function reaches the minimum value: After performing a linear fit, the coefficient of determination R 2 , used to evaluate the fitting effect of the linear model, R 2 It indicates the degree of explanation of the model to the data, and its calculation formula is: Where T i is the actual temperature data point, is the temperature value predicted by the regression model, is the average value of the actual temperature data; R 2 The value range is between 0 and 1, indicating the degree of fit of the linear model to the data; the closer the value is to 1, the better the model fits the data, and the stronger the linear relationship between the data; the closer the value is to 0, the poorer the model's ability to explain the data, and the data presents greater volatility and randomness; A threshold value (set to 0.75 in this embodiment) is preset to evaluate the regression coefficient R 2 , which is based on the analysis of historical data of the sensor under normal working conditions and represents the reasonable range of linear change of temperature data. 2 Compared with the threshold of 0.75, if R 2 <0.75, indicating that the linear relationship between temperature data is weak, and there may be large fluctuations or randomness, which usually indicates that the temperature sensor may be faulty, causing abnormal data fluctuations. At this time, the system outputs a prompt message "Need to check whether the sensor is faulty, causing data fluctuations" and terminates the current diagnostic process. 2 ≥0.75, indicating that there is a strong linear relationship between the temperature data, indicating that the working state of the sensor is relatively stable, and the temperature change conforms to the expected linear trend. At this time, it is considered that the sensor is operating normally and the diagnosis process is completed. 2 If the value always hovers near the threshold, other analysis methods such as polynomial regression or time series analysis can be introduced to further verify the accuracy of the sensor readings and ensure the comprehensiveness and accuracy of fault diagnosis.

[0033] This embodiment adopts a multi-level fault diagnosis mechanism. Through a comprehensive process from data collection, data verification, incremental analysis, smoothing processing to regression analysis, it conducts a detailed and comprehensive investigation of various possible faults of the temperature sensor, thereby ensuring the accuracy and reliability of the diagnosis results. It can realize multi-level fault diagnosis of the temperature sensor and effectively improve the accuracy and reliability of sensor fault detection.

[0034] Although this embodiment is directed to a temperature sensor, it is also applicable to other types of environmental sensors, such as humidity, pressure, light intensity, etc. By adjusting the corresponding diagnostic parameters and analysis methods, it can be widely used in various environmental monitoring systems and has good versatility.

Claims

1. A sensor fault diagnosis method, characterized in that the steps include: Step S1: Acquire data of target environment sensors and define the normal working range of the sensors, including threshold range and data change characteristics; Read the collected data T of the sensor in real time and record the data as the sensor reading at the current moment; Step S2: Determine whether the obtained sensor reading is within a preset threshold range; when the sensor reading is not within the preset range, output an abnormal warning, indicating that there may be a sensor communication failure or a sensor hardware failure; otherwise, continue to analyze the data change trend.

2. The sensor fault diagnosis method according to claim 1, characterized in that: Also includes: Step S3: Calculate the sensor reading increments ΔT at adjacent time points, construct the increment distribution histogram of multiple time points, count the frequencies of different increment intervals, and extract the increment distribution feature values, such as the 95% and 100% increment thresholds (ΔX 95 and ΔX 100 ); Step S4: Determine the ratio relationship of the incremental distribution characteristic values, i.e. ΔX 95 / ΔX 100 Whether the preset conditions are met; if the ratio is less than the preset value, an abnormal warning is output, indicating that the sensor may have position abnormality, occlusion or other external interference problems; otherwise, continue to perform subsequent fault diagnosis steps; Step S5: Perform sliding average processing on all sensor readings to eliminate obviously abnormal fault data; Step S6: Perform regression analysis on the smoothed data and calculate the regression coefficient R 2 To evaluate the strength of the linear relationship between data; R 2 The value is compared with the preset regression coefficient preset value; the regression coefficient R is determined 2 Is it less than the preset value? If it is less than the preset value, an abnormal warning is output, indicating that the sensor may be faulty, resulting in abnormal data fluctuations; otherwise, the output indicates that the sensor is operating normally.

3. The sensor fault diagnosis method according to claim 2, characterized in that: The sensor is a temperature sensor, a humidity sensor, a pressure sensor or a light intensity sensor.

4. The sensor fault diagnosis method according to claim 2, characterized in that: The incremental distribution characteristic values ​​are found from multiple angles, including the median, mode and mean, which are used as the center points of the data distribution, and then the width of the 95% ratio is calculated.

Images