A sensor fault detection method and apparatus for a structural health monitoring system

Abstract

The application discloses a kind of sensor fault detection method and device of structural health monitoring system, it is related to structural health monitoring technical field, comprising: obtaining the nonlinear structure monitoring data of sensor system collected by structural health monitoring system, nonlinear structure monitoring data is handled using improved geometric PNL hybrid model, obtain linear to-be-separated mixed signal, using FastICA model to process to-be-separated mixed signal, obtain multiple independent elements, determine the separation matrix of improved geometric post-nonlinear independent component analysis model according to multiple independent elements, using the processing of improved geometric post-nonlinear independent component analysis to real-time collection nonlinear structure monitoring data, determine whether sensor system exists fault and the sensor of fault occurrence.The method can still complete linearization processing to sensor mixed signal under the condition that prior knowledge is unknown, compared with simple linear ICA analysis algorithm, it is more suitable for complex nonlinear structure.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology, and more specifically to a sensor fault detection method and apparatus for a structural health monitoring system. Background Technology

[0002] In recent years, for some large public buildings, such as long-span buildings and super high-rise buildings, conventional structural testing is no longer sufficient to meet the requirements of structural safety, necessitating structural health monitoring (SHM). An SHM system typically consists of three main subsystems: a sensor subsystem, a data transmission subsystem, and a health assessment subsystem. Structural response is first measured by the sensor subsystem and then transmitted through the data transmission subsystem; therefore, the accuracy and reliability of the structural health assessment results largely depend on the type, quantity, and quality of the data measured by the sensor subsystem. Current research on SHM systems mainly focuses on structural condition assessment and optimized sensor placement, with less attention paid to sensor fault diagnosis. Factors such as sensor manufacturing quality, harsh operating environments (electromagnetic interference, noise, temperature, humidity, etc.), and performance degradation can all lead to sensor failure. To ensure the normal operation of the SHM system, sensors and other hardware must be replaceable without affecting data continuity. Therefore, online monitoring and real-time fault diagnosis of the sensor subsystem in SHM systems are of great significance.

[0003] Sensor fault detection is used to determine whether a fault has occurred in a sensor network. Establishing accurate mathematical models for sensors is extremely difficult due to various interferences, including environmental factors. Independent Component Analysis (ICA), however, does not require an accurate mathematical model and is suitable for analyzing non-Gaussian systems such as long-span or high-rise buildings. While ICA is effective in detecting certain types of sensor faults, its effectiveness is limited when dealing with typical sensor faults, such as the difficulty in determining the influencing factors of sensor faults in SHM (Sensitive Component Analysis) systems. Nguyen et al. proposed a novel nonlinear ICA algorithm—Geometric Post Nonlinear ICA (gpICA)—which can linearize mixed signals without any prior information or assumptions, making it suitable for detecting sensor faults in nonlinear monitoring data.

[0004] Sensor fault isolation refers to identifying specific faulty sensors in a sensor network. Kerschen et al. divided the sensor network into normal sensor groups and faulty sensor groups. By assuming each sensor is a missing variable in turn, they calculated the corresponding fault detection index by removing the missing variables from the normal and reference sensor groups. If the calculated fault detection index changes the most compared to the case without removing the missing variables, then the sensor whose missing variable was removed, i.e., the sensor being detected, is identified as a faulty sensor. However, in large-scale structural SHM applications, multiple sensors may fail. If this method is used to test each possible sensor combination sequentially, the fault isolation process will be very time-consuming. Probabilistic quantification methods can quantify the failure probability of each sensor by defining a sensor fault index. Based on the failure probability, the isolability of each faulty sensor in the sensor network can be studied, greatly improving the efficiency of fault isolation. Although this method can quantify the failure probability of each sensor, it is only applicable when the environmental noise is Gaussian distributed. Sensor measurements in long-span or super high-rise building SHM systems usually do not follow a Gaussian distribution, while the ICA algorithm is suitable for diagnosing sensor faults in the system because it is applicable to non-Gaussian systems. However, most current ICA-based sensor fault diagnosis methods require the establishment of accurate mathematical models. Summary of the Invention

[0005] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, a first aspect of this invention proposes a sensor fault detection method for a structural health monitoring system, comprising:

[0006] Acquire nonlinear structural monitoring data collected by the sensor system of the structural health monitoring system;

[0007] An improved geometric PNL mixture model is used to process nonlinear structural monitoring data to obtain a linear mixture signal to be separated. The FastICA model is then used to process the mixture signal to obtain multiple independent elements. Based on these independent elements, the separation matrix of the improved geometric post-nonlinear independent element analysis model is determined. In the improved geometric PNL mixture model, the time exponent of the mixture is used to... t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points in the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of the two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane.

[0008] An improved geometric post-nonlinear independent element analysis model is used to process real-time acquired nonlinear structural monitoring data to determine whether the sensor system has faults and which sensors have failed.

[0009] Furthermore, through the time index of the mixture tThe γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points on the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of these two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane, including:

[0010] Randomly determine a reference surface and a surface to be moved. Determine the coordinates of any two points and the coordinates of the first midpoint of any two points on the reference surface, where the reference surface is any surface.

[0011] Time index of the mixture t Determine the coordinates of two accompanying points on the surface to be moved and the coordinates of the second midpoint of the line connecting the two accompanying points for any two points in the reference plane within the same sampling time.

[0012] Using a preset first formula to process the coordinates of any two points, the coordinates of two accompanying points, and the coordinates of the first midpoint, the first γ quadrant coordinate value of the accompanying point coordinates of the first midpoint on the surface to be moved is obtained. Using a preset second formula to process the first γ quadrant coordinate value and the second midpoint coordinate, the second γ quadrant coordinate value of the accompanying point of the first midpoint after the move is obtained.

[0013] The expression (1) of the first preset formula is:

[0014]

[0015] in, This represents the γ quadrant coordinate value of the first midpoint between any two points on the reference plane.

[0016] This represents the γ quadrant coordinate value of one of any two points on the reference plane;

[0017] This represents the γ quadrant coordinate value of the other point among any two points on the reference plane.

[0018] This represents the γ quadrant coordinate value of one of the two accompanying points on the surface to be moved;

[0019] This represents the γ quadrant coordinate value of the other point among the two accompanying points on the surface to be moved.

[0020] The expression (2) of the second preset formula is:

[0021] in, Points on the surface to be moved p c The γ quadrant coordinates before movement; Points on the surface to be moved p c The shifted γ quadrant coordinates For learning rate, The range of values ​​is 0 < <1;

[0022] The γ quadrant coordinates of the accompanying point of the first midpoint coordinates after and before the move are processed using the third preset formula to obtain the error value. If the error value is greater than the preset first threshold, the γ quadrant coordinates of the accompanying point of the first midpoint coordinates after the move are iteratively calculated.

[0023] The expression (3) of the three preset formulas is:

[0024]

[0025] In the formula N k This represents the number of points to be updated in each iteration, and n represents the number of observed signals; The γ quadrant coordinates of the accompanying point represent the coordinates of the first midpoint after the movement. The γ quadrant coordinates of the accompanying point represent the coordinates of the first midpoint before the movement.

[0026] Furthermore, if the error value is less than the preset first threshold, the iteration ends, and the γ quadrant coordinate values ​​of all the moved first midpoint coordinates generated by the iteration are sorted in ascending order and assigned to each signal to be separated.

[0027] The fourth preset formula is used to perform a smooth fit on each signal to be separated to obtain the signal function to be separated;

[0028] The expression (4) of the fourth preset formula is:

[0029]

[0030] z i s The signal to be separated L The time window length is defined as a value within the range of (0, 1). N () is an integer, where N is the number of samples;

[0031] The FastICA model is used to process the signal function to be separated, and multiple independent elements representing the independent signals are obtained. The separation matrix is ​​then determined based on these independent elements.

[0032] Furthermore, the separation matrix is ​​determined based on multiple independent elements, including:

[0033] By using the Euclidean norm to process all independent elements, we obtain multiple independent elements arranged in sequence;

[0034] By using the top-ranked independent elements as the dominant independent elements, a separation matrix is ​​obtained.

[0035] Furthermore, the improved geometric post-nonlinear independent element analysis model is used to process the real-time acquired nonlinear structural monitoring data to determine whether there is a fault in the sensor system, including:

[0036] The Euclidean norm is used to process the separation matrix to obtain the separation principal part matrix and the cofactor matrix;

[0037] The independent element vectors of the principal part and the independent element vector of the remainder are calculated by multiplying the principal part and the remainder of the separation matrix with the nonlinear data acquired at any time.

[0038] The independent element vectors of the main part, the independent element vectors of the remainder, and the monitoring data during the operation of the sensor system are processed according to the fifth preset formula to obtain the normal monitoring statistics and real-time monitoring statistics. The kernel density estimation algorithm is used to process the normal monitoring statistics to obtain the normal monitoring statistics threshold.

[0039] The fifth preset formula includes I 2 Expression for monitoring statistics (5) I e 2 The expression for the monitoring statistics (6) and SPE The expression for the monitoring statistics (7);

[0040]

[0041]

[0042]

[0043] In the formula, , , Represents the principal matrix. This represents the monitoring data during system operation. Represents the remainder matrix. Denotes the independent element vectors of the principal component. Denotes the independent element vectors of the remainder. ,in, This is the whitening matrix;

[0044] If the real-time monitoring statistic value is greater than the normal monitoring statistic threshold, then the sensor system is determined to be faulty; otherwise, the sensor system is not faulty.

[0045] Furthermore, the improved geometric post-nonlinear independent element analysis model is used to process the real-time acquired nonlinear structural monitoring data to identify faulty sensors in the sensor system, including:

[0046] Contribution analysis algorithms are used to process real-time monitoring statistics. The contribution analysis formula includes:

[0047] I 2 The expression for the contribution value (8) I e 2 The expression for the contribution value (9) SPE The expression for the contribution value (10):

[0048]

[0049]

[0050]

[0051] In the formula, express I 2 The contribution value; express I e 2 The contribution value; express SPE The contribution value.

[0052] Furthermore, after identifying the faulty sensor in the sensor system, the process also includes:

[0053] The sensor fault type is determined using a mathematical model of sensor faults. The sensor fault types include: fixed deviation fault, linear deviation fault, constant gain fault, accuracy degradation fault, jamming fault, white noise jamming fault, and zero line drift fault.

[0054] Another aspect of the present invention provides a sensor fault detection device for a structural health monitoring system, comprising:

[0055] The data acquisition module is used to acquire nonlinear structural monitoring data collected by the sensor system of the structural health monitoring system;

[0056] The model determination module is used to process nonlinear structural monitoring data using an improved geometric PNL mixture model to obtain a linear mixture signal to be separated. The FastICA model is then used to process the mixture signal to be separated, obtaining multiple independent elements. Based on these independent elements, the separation matrix of the improved geometric post-nonlinear independent element analysis model is determined. In the improved geometric PNL mixture model, the time exponent of the mixture is used to determine the separation matrix. tThe γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points in the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of the two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane.

[0057] The fault determination module is used to process real-time acquired nonlinear structural monitoring data using an improved geometric post-nonlinear independent element analysis model to determine whether the sensor system has a fault and which sensor has a fault.

[0058] In another aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement a sensor fault detection method for a structural health monitoring system as described in any of the first aspects.

[0059] In another aspect, the present invention provides a computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the sensor fault detection method of the structural health monitoring system according to any one of the first aspects.

[0060] This invention provides a sensor fault detection method for a structural health monitoring system, which has the following advantages compared with the prior art:

[0061] 1. This invention improves the gpICA algorithm, enabling its application in structural health monitoring systems when the source signal is unknown. The improved gpICA algorithm considers the nonlinear ICA problem from a multidimensional perspective, and can still linearize mixed signals even when prior knowledge is unknown. Compared to simple linear ICA analysis algorithms, it is more suitable for complex nonlinear structures, providing theoretical support for sensor fault diagnosis in structural health monitoring systems.

[0062] 2. The improved gpICA-based sensor fault detection method established in this invention improves fault detection efficiency by dividing the independent element space and selecting an appropriate number of independent elements. Sensor fault detection is achieved by comparing monitoring statistics with thresholds, and the applicability of this method is studied through numerical simulation. It can be seen that this algorithm has a higher fault detection rate than traditional linear ICA fault detection.

[0063] 3. The improved gpICA-based sensor isolation method established in this invention, after separating the monitoring data of the structural health monitoring system using the improved gpICA algorithm, makes the contribution of monitoring statistics more sensitive to faulty sensors. In numerical simulations, it can accurately locate the faulty sensor in the sensor system when one or multiple sensors fail. The results show that the improved gpICA fault isolation method is more suitable for faulty sensor isolation in structural health monitoring systems. Attached Figure Description

[0064] To more clearly illustrate the technical solutions of the present invention, the accompanying drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0065] Figure 1 This is a flowchart provided by the present invention;

[0066] Figure 2 This invention provides a nonlinear model mixing and unmixing process;

[0067] Figure 3 This is a geometric description of the PNL mixing and separation system in three-dimensional space provided by the present invention;

[0068] Figure 4 This is a three-dimensional spatial representation of the geometric PNL hybrid separation system provided by the present invention;

[0069] Figure 5 This is a three-dimensional spatial representation of the geometric PNL hybrid separation system provided by the present invention;

[0070] Figure 6 This is a mathematical model diagram of sensor faults provided by the present invention;

[0071] Figure 7 This is a diagram showing the position of the acceleration sensor provided by the present invention;

[0072] Figure 8 This is the monitoring statistic provided by the present invention under operating condition 1;

[0073] Figure 9 These are the monitoring statistics provided by this invention under operating condition 2;

[0074] Figure 10 This invention provides an accelerometer sensor position diagram;

[0075] Figure 11 This invention provides contribution values ​​for various monitoring statistics under operating condition 1;

[0076] Figure 12This invention provides contribution values ​​for various monitoring statistics under operating condition 2;

[0077] Figure 13 This is a schematic diagram of the device structure provided by the present invention. Detailed Implementation

[0078] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] This specification provides the operational steps for the methods described in the embodiments or flowcharts, but may include more or fewer operational steps based on conventional or non-inventive labor. In actual system or server product execution, the methods shown in the embodiments or drawings may be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0080] Related knowledge:

[0081] Suppose there exists a set of unknown source signals. Then the observed signal It is the source signal linear mixtures, t For time variables, the linear mixed model is shown in the following equation:

[0082]

[0083] In the formula, ; ; for Composition Non-singular mixture matrix.

[0084] Taleb and Jutten proposed the PNL mixture model by adding a nonlinear distortion function to the linear mixture model. Achieving nonlinear mixing. The nonlinear mixing process is shown in the following equation:

[0085]

[0086] In the formula, These are nonlinear distortion functions that are independent and invertible.

[0087] The reverse process of mixing is the separation process, as shown in the following equation:

[0088]

[0089] In the formula, yes nonlinear inverse function, yes The estimated value.

[0090] The principle of mixing and unmixing in the PNL model is as follows: Figure 2 As shown, the left side represents the mixing process; the right side represents the separation process, which is the reverse of the mixing process.

[0091] For PNL mixture models, constraints must be imposed on their parameters to obtain a unique solution, while geometric PNL mixture models do not require any additional assumptions and can effectively handle nonlinear distortion. In three-dimensional space, the data graph of a linear mixture is planar, while the data graph of a nonlinear mixture is a curved surface. Therefore, the approximation process of transforming a nonlinear mixture into a linear mixture is to transform a curved surface into a planar one. The geometric description of a PNL mixture separation system in three-dimensional space is as follows: Figure 3 As shown.

[0092] (1) Plane determination method: such as Figure 4 As shown, , It is a curved surface in three-dimensional space Any two points above, for , A point on a line is a vertex if and only if for any... All fall on the surface When above, curved surface It is a plane.

[0093] (2) Method for converting a curved surface into a plane: repeatedly move the curved surface For each point on the plane, move it to the plane. The corresponding position on the graph. Before performing the conversion, it is necessary to understand the concept of paired points.

[0094] Define a pair of adjacent points: for any pair of points , When having the same , Coordinates, then for The point of companionship, point to point Call them companions.

[0095] Define b as an assumption , For two planes in space, , For plane Any two points on, , For plane Any two points on, and and If they are paired points, then... , Midpoint of the line connecting two points and , Midpoint of the line connecting two points It must be a pair of companions.

[0096] Suppose there exists a plane The values ​​of all points on it ( The values ​​of the axes are known. First, on the surface... Choose any two points , ,Sure , Midpoint of the line Then, on the plane The above determined , companion points , ,calculate , Midpoint of the line On curved surfaces Found on companion points and put Transform into In each iteration, a reference plane is randomly selected. This is used to transform the corresponding surface, ultimately linearizing the data. The process of moving a point on the surface to the plane is as follows: Figure 4 As shown, Figure 4 (a) is Before the point moves, Figure 4 (b) is After the point is moved.

[0097] The linearized mixed signal was obtained through the geometric PNL mixture model. Next, linear separation is performed using the linear ICA algorithm to obtain the original signal features. The Fast Independent Component Analysis (FastICA) algorithm within the linear ICA algorithm is widely used due to its simple computation and fast convergence. This invention adopts the FastICA algorithm based on negative entropy. According to the central limit theorem, if a random variable... Consists of many independent random variables Composition, as long as If the mean and variance are of the same order of magnitude, then the random variable... compared to It is closer to a Gaussian distribution, which also means that independent random variables Compared to random variables Gaussian variables exhibit higher non-Gaussianity, thus non-Gaussianity represents independence. Information theory states that among all random variables with equal variance, Gaussian variables have the highest entropy, meaning entropy can be used to measure non-Gaussianity.

[0098] To obtain a measure of the non-Gaussianity of Gaussian variables that are zero and always non-negative, the normalized form of differential entropy, called negative entropy, is usually used. Hyvarinen, a foreign scholar, proposed an approximate formula for calculating negative entropy, as shown below:

[0099]

[0100] In the formula, This is for calculating the mean. —Nonlinear function.

[0101] like Figure 1 As shown, a sensor fault detection method for a structural health monitoring system includes the following steps:

[0102] Step 101: Acquire nonlinear structural monitoring data collected by the sensor system of the structural health monitoring system;

[0103] Step 102: Train the improved geometric post-nonlinear independent element analysis algorithm using the training set to obtain multiple independent elements, and use these independent elements to determine the separation matrix; in the improved geometric post-nonlinear independent element analysis algorithm, the time exponent of the mixture is used... t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points in the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of the two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane.

[0104] Specifically, in step 102, the time index of the mixture is... t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points on the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of these two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane, including:

[0105] Randomly determine a reference surface and a surface to be moved. Determine the coordinates of any two points and the coordinates of the first midpoint of any two points on the reference surface, where the reference surface is any surface.

[0106] Time index of the mixture tDetermine the coordinates of two accompanying points on the surface to be moved and the coordinates of the second midpoint of the line connecting the two accompanying points for any two points in the reference plane within the same sampling time.

[0107] Using a preset first formula to process the coordinates of any two points, the coordinates of two accompanying points, and the coordinates of the first midpoint, the first γ quadrant coordinate value of the accompanying point coordinates of the first midpoint on the surface to be moved is obtained. Using a preset second formula to process the first γ quadrant coordinate value and the second midpoint coordinate, the second γ quadrant coordinate value of the accompanying point of the first midpoint after the move is obtained.

[0108] The expression (1) of the first preset formula is:

[0109] The expression (2) of the second preset formula is:

[0110] in, Points on the surface to be moved The γ quadrant coordinates before movement; Points on the surface to be moved The shifted γ quadrant coordinates For learning rate, The range of values ​​is 0 < <1;

[0111] The γ quadrant coordinates of the accompanying point of the first midpoint coordinates after and before the move are processed using the third preset formula to obtain the error value. If the error value is greater than the preset first threshold, the γ quadrant coordinates of the accompanying point of the first midpoint coordinates after the move are iteratively calculated.

[0112] The expression (3) of the three preset formulas is:

[0113]

[0114] In the formula This indicates the number of points to be updated in each iteration. Indicates the number of observed signals; The γ quadrant coordinates of the accompanying point represent the coordinates of the first midpoint after the movement. The γ quadrant coordinates of the accompanying point represent the coordinates of the first midpoint before the movement.

[0115] If the error value is less than the preset first threshold, the iteration ends, and the γ quadrant coordinate values ​​of all the moved first midpoint coordinates generated by the iteration are sorted in ascending order and assigned to each signal to be separated.

[0116] The fourth preset formula is used to perform a smooth fit on each signal to be separated to obtain the signal function to be separated;

[0117] The expression (4) of the fourth preset formula is:

[0118]

[0119] The signal to be separated The time window length is defined as a value within the range of (0, 1). integers, The number of samples;

[0120] The FastICA model is used to process the signal function to be separated, and multiple independent elements representing the independent signals are obtained. The separation matrix is ​​then determined based on these independent elements.

[0121] All independent elements are processed using the Euclidean norm to obtain multiple independent elements arranged in order; the independent elements with the highest order are used as the dominant independent elements to obtain the separation matrix.

[0122] Specifically, a reference plane is randomly selected. Randomly select the surface ;

[0123] In the reference plane Randomly select two points , , ; Utilizing the time index t On the curved surface Γ xi Find their companion points , , Calculate the point using equation (1) of value;

[0124] in, This represents the γ quadrant coordinate value of the first midpoint between any two points on the reference plane.

[0125] This represents the γ quadrant coordinate value of one of any two points on the reference plane;

[0126] This represents the γ quadrant coordinate value of the other point among any two points on the reference plane.

[0127] This represents the γ quadrant coordinate value of one of the two accompanying points on the surface to be moved;

[0128] This represents the γ quadrant coordinate value of the other point among the two accompanying points on the surface to be moved.

[0129] Calculate using equation (2) Since the first two coordinates are unknown, it cannot be determined. Is it located in , Connect the points. Therefore, the converse of the plane determination does not always hold, meaning there may be more than one point that satisfies definition b but is not. The companion point. There is currently no way to completely solve this problem, but it can be addressed by applying a learning rate. It can eliminate some errors in the conversion process. The range of values ​​is 0 < <1. By doing so, the transformation will take longer, but it will converge stably to a plane;

[0130] The error is calculated using equation (3), where It is the number of points to be updated in each iteration, if the error... Return to the reference plane Two points are randomly selected again to proceed with the subsequent steps. This is the threshold for stopping linearization. Otherwise, the linearization process stops and proceeds to the next step.

[0131] Will The value assigned to ,Will Sort in ascending order to get Using equation (4) to analyze the signal Smoothness optimization is performed, where The time window length is defined as a value within the range of (0, 1). integers, The number of samples;

[0132] The linear separation algorithm FastICA is used for The separation signal is obtained. .

[0133] After separating the structural monitoring data, the improved gpICA algorithm yields many independent elements. As can be seen from the basic principles of ICA, the number of separated source signals... Greater than or equal to the actual number of source signals To select an appropriate number of independent variables, all independent variables should first be arranged in a suitable order, and then the first few should be selected. Each independent element, as the dominant independent element, utilizes the preceding... Establish a separation matrix with each independent element as the dominant independent element. .

[0134] By dividing the nonlinear structural monitoring data collected by the sensor system of the acquired structural health monitoring system into a training set and a test set, and continuously performing the above steps using the training set, a trained separation matrix is ​​obtained. In the separation matrix Input test data to verify the separation matrix The correctness of the statement.

[0135] Step 103: Calculate the nonlinear structure monitoring data collected by the sensor system using the separation matrix to obtain the normal monitoring statistics value and the real-time monitoring statistics value, and estimate the first monitoring statistics threshold using the kernel density estimation algorithm. Based on the comparison result between the real-time monitoring statistics value and the first monitoring statistics threshold, determine whether there is a fault in the sensor system.

[0136] In step 103, the separation matrix is ​​processed using the Euclidean norm to obtain the separation principal matrix and the cofactor matrix;

[0137] The independent element vectors of the principal part and the independent element vector of the remainder are calculated by multiplying the principal part and the remainder of the separation matrix with the nonlinear data acquired at any time.

[0138] The independent element vectors of the main part, the independent element vectors of the remainder, and the monitoring data during the operation of the sensor system are processed according to the fifth preset formula to obtain the normal monitoring statistics and real-time monitoring statistics. The kernel density estimation algorithm is used to process the normal monitoring statistics to obtain the normal monitoring statistics threshold.

[0139] The fifth preset formula includes Expression for monitoring statistics (5) The expression for the monitoring statistics (6) and SPE The expression for the monitoring statistics (7);

[0140]

[0141]

[0142]

[0143] In the formula, , , Represents the principal matrix. This represents the monitoring data during system operation. Represents the remainder matrix. Denotes the independent element vectors of the principal component. Denotes the independent element vectors of the remainder. ,in, For whitening matrix, ;

[0144] If the real-time monitoring statistic value is greater than the normal monitoring statistic threshold, it is determined whether there is a fault in the sensor system; otherwise, the sensor system is not faulty.

[0145] Specifically, this invention will use the Euclidean norm - L2 norm to separate the matrix. W The row vectors are sorted, and the magnitude of the row vectors represents the proportion of mixed signal contained in this independent element. The separation matrix is ​​calculated... W Sort the row vectors by their L2 norm percentages and select... d individual principal W d As the main part, and W Remove W d The remaining matrix W e As a remainder. For new data at a certain moment. x new Based on the main part W d and the remaining part W e Independent element vectors can be calculated, and then the monitoring statistics of the monitoring data during normal operation of the sensor system can be calculated according to formulas (5), (6), and (7). The kernel density estimation method is used to estimate the threshold of the monitoring statistics. The monitoring statistics of the real-time monitoring data are compared with the threshold. If a large number of points exceed the threshold, the sensor system is faulty. This invention is based on I 2 , I e 2 and SPE Three monitoring statistics are used for sensor fault detection. I 2 Monitoring statistics are used to characterize changes within the model. SPE Monitoring statistics are used to characterize changes in the out-of-model residuals. I e 2 Monitoring statistics can compensate the main body W d Errors that occur when an incorrect quantity is selected.

[0146] Step 104: When the sensor system is faulty, the contribution analysis algorithm is used to process the second monitoring statistic to obtain the contribution value of the real-time monitoring statistic and draw a contribution value histogram. Based on the contribution value histogram, the faulty sensor is identified.

[0147] In step 104, the second monitoring statistic is processed using the contribution analysis formula to obtain the contribution value of the real-time monitoring statistic, and a contribution value histogram is plotted. Based on the contribution value histogram, the faulty sensor is determined, including:

[0148] Contribution analysis algorithms are used to process real-time monitoring statistics. The contribution analysis formula includes:

[0149] I 2 The expression for the contribution value (8) I e 2 The expression for the contribution value (9) SPE The expression for the contribution value (10):

[0150]

[0151]

[0152]

[0153] In the formula, express I 2 The contribution value; express I e 2 The contribution value; express SPE The contribution value.

[0154] Specifically, when a fault is determined in the sensor system, it is necessary to locate the specific faulty sensor. The key method for fault isolation using the gpICA model is contribution analysis, where the sensor with the largest contribution value of the monitoring statistics is the faulty sensor.

[0155] Calculate each variable I 2 Statistics I e 2 Statistics and SPE The contribution values ​​of the statistics are calculated and plotted into a histogram, i.e., a contribution graph. Faulty sensors will produce a larger contribution compared to normal sensors. I 2 value, I e 2 Value and SPE These values ​​will be detected, and the corresponding faulty sensors will be isolated. The relative size of the contribution plot represents the contribution of each variable to the prediction error, or the degree of mismatch between the sample and the model. In some cases, especially when statistics are constantly changing, the average contribution of variables over a period of time should be used.

[0156] In summary, the improved gpICA sensor fault isolation method, which involves calculating... I2 Statistics I e 2 Statistics and SPE The contribution values ​​of the statistics are calculated and a histogram is plotted. The sensor with the largest contribution value is the faulty sensor.

[0157] In one possible implementation, the sensor fault type is determined using a sensor fault mathematical model, and the sensor fault type includes: fixed deviation fault, linear deviation fault, constant gain fault, accuracy degradation fault, jamming fault, white noise jamming fault, and zero line drift fault.

[0158] In the embodiments provided by this invention, Kullaa summarized seven types of sensor faults: fixed deviation, linear deviation, constant gain, accuracy degradation, jamming, white noise jamming, and zero-line drift. The first four types of sensor faults are partial sensor failures, commonly referred to as soft sensor faults; the latter three types are complete sensor failures, commonly referred to as hard sensor faults. This paper will use the above mathematical models of sensor faults to simulate faulty sensors and verify the effectiveness of the proposed sensor fault and isolation method.

[0159] make z *( t )represent t The true value of the measured variable at any given time. w ( t () represents free noise. Assuming that the free noise follows a normal distribution with a mean of zero, the mathematical expressions for sensor faults are shown in Table 1, where... a , b , G For fixed values, c The intercept is... f The slope is the parameter used to control the fault amplitude of the corresponding sensor. e(t) It is random noise with a mean of zero that causes sensor failure.

[0160] Table 1 Mathematical Expressions for Sensor Faults

[0161]

[0162] Sensor partial failure fault model, such as Figure 6 As shown, Figure 6 (a), (b), (c), and (d) represent fixed deviation fault, linear deviation fault, constant gain fault, and accuracy degradation fault, respectively. As can be seen from the figure, some of the failed data can still accurately reflect the actual situation of the structure after correction. Therefore, this paper mainly focuses on some failed sensors for sensor fault diagnosis. Specific implementation examples:

[0164] The bridge structure is a typical long-span structure. This invention uses MATLAB software to establish a three-span continuous concrete beam model. The model dimensions are: beam length 8m, cross-section 0.6m × 0.6m rectangular section, and the model's elastic modulus is... E 3×10 7 kN / m 2 Poisson's ratio μ Density 0.2 ρ 2500 kg / m 3 The entire beam is divided into 80 units along its longitudinal direction, each unit being 0.1m in length. Random loads are used as the excitation for the continuous beam, and the Newmark-β method is used to calculate the acceleration time history response of the continuous beam. Assume that the accelerometers are placed at distances of 0.6m, 1.2m, 3m, and 4m from the left end support of the beam, and are numbered 1, 2, 3, and 4 respectively. Figure 7 As shown.

[0165] Using MATLAB software, a total of 600 random loads were generated as excitations applied to the structure. Each accelerometer generated a total of 600 acceleration values. The first 300 structural response data were used for gpICA training, and the 301st to 600th data were used for sensor fault detection under fault conditions.

[0166] Assuming the bridge structure is intact, and only the sensors malfunction, there are two scenarios: a single sensor malfunction or multiple sensor malfunctions. Two operating conditions are set up: Condition 1 is when only sensor 4 experiences an accuracy degradation fault, while all other sensors are normal; Condition 2 is when sensor 1 experiences a linear deviation fault, and sensor 3 experiences a constant gain fault. The operating condition table is shown in Table 2.

[0167] Table 2 Sensor Operating Conditions

[0168]

[0169] Under operating condition 1, the control limits for three monitoring statistics were determined using the kernel density estimation method. The results after gpICA test are shown in Figure 8. Although I 2 More than 30% of the statistical values ​​exceeded the limit, but I e 2 Statistics and SPE The fact that over 90% of the statistical points exceeded the limits indicates that the sensor was faulty during the process.

[0170] Under operating condition 2, the control limits for three monitoring statistics were determined using the kernel density estimation method, and the results after gpICA test are as follows: Figure 9 As shown, I 2 StatisticsI e 2 Statistics and SPE The fact that over 85% of the statistical points exceeded the limits indicates that the sensor malfunctioned during the process.

[0171] from Figure 8 and Figure 9 As can be seen from the fault detection diagram, the sensor fault detection method based on the improved gpICA is accurate in detecting sensors. A large number of points of the three monitoring statistics of the faulty sensor exceed the limit, indicating that the three monitoring statistics are sensitive to sensor faults. When one or more sensors fail, the faulty sensors in the sensor system can be effectively detected.

[0172] To ensure a clear effect during sensor fault isolation, a total of 10 sensors were simulated. Assume the accelerometers are positioned at distances of 0.6m, 1.2m, 1.8m, 3.2m, 4m, 5m, 5.8m, 6.4m, 7m, and 7.6m from the left end support of the bridge, and are numbered 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10 respectively. Figure 10 As shown.

[0173] Assuming the bridge structure is intact, and only the sensors malfunction, there are two scenarios: a single sensor malfunction or multiple sensor malfunctions. Two operating conditions are set up: Condition 1 is when only sensor 5 experiences a linear deviation fault, while the other sensors are normal; Condition 2 is when sensor 3 experiences a fixed deviation fault, sensor 7 experiences an accuracy degradation fault, while the other sensors are normal. The operating condition table is shown in Table 3.

[0174] Table 3 Sensor Operating Conditions

[0175]

[0176] Under operating condition 1, the contribution graph of the statistics is as follows: Figure 11 As shown, it can be seen that all statistics are highest for sensor number 5, which indicates that sensor number 5 is the faulty sensor.

[0177] Under operating condition 2, the contribution graph of the statistics is as follows: Figure 12 As shown, the analysis of the contribution of each sensor to the sensor system failure indicates that sensors 3 and 7 have the highest monitoring statistics, thus identifying sensors 3 and 7 as the faulty sensors. The analysis also shows that the gpICA-based fault isolation method is applicable when multiple sensors fail simultaneously.

[0178] from Figure 11 and Figure 12As can be seen from the contribution plot, after data processing based on the improved gpICA algorithm, the statistical contribution values ​​more accurately reflect sensor faults. The contribution plot can be used to accurately locate faulty sensors, achieving sensor isolation requirements.

[0179] This invention provides a sensor fault detection device 200 for a structural health monitoring system, such as... Figure 13 As shown, it includes:

[0180] Data acquisition module 201 is used to acquire nonlinear structural monitoring data collected by the sensor system of the structural health monitoring system;

[0181] The model determination module 202 is used to process nonlinear structural monitoring data using an improved geometric PNL mixture model to obtain a linear mixture signal to be separated. It then uses a FastICA model to process the mixture signal to be separated, obtaining multiple independent elements. Based on these independent elements, it determines the separation matrix of the improved geometric post-nonlinear independent element analysis model. In the improved geometric PNL mixture model, the separation matrix is ​​determined through the time exponent of the mixture. t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points in the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of the two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane.

[0182] The fault determination module 203 is used to process real-time acquired nonlinear structural monitoring data using an improved geometric post-nonlinear independent element analysis model to determine whether the sensor system has a fault and which sensor has a fault.

[0183] In another embodiment of the present invention, a device is also provided, the device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to implement the sensor fault detection method of the structural health monitoring system described in the embodiment of the present invention.

[0184] In another embodiment of the present invention, a computer-readable storage medium is also provided, wherein at least one instruction, at least one program, code set or instruction set is stored in the storage medium, wherein the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the sensor fault detection method of the structural health monitoring system described in the embodiment of the present invention.

[0185] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes multiple computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates multiple available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0186] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0187] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0188] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A sensor fault detection method for a structural health monitoring system, characterized in that, include: Acquire nonlinear structural monitoring data collected by the sensor system of the structural health monitoring system; An improved geometric PNL mixture model is used to process nonlinear structural monitoring data to obtain a linear mixture signal to be separated. The FastICA model is then used to process the mixture signal to obtain multiple independent elements. Based on these independent elements, the separation matrix of the improved geometric post-nonlinear independent element analysis model is determined. In the improved geometric PNL mixture model, the time exponent of the mixture is used to... t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points in the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of the two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane. An improved geometric post-nonlinear independent element analysis model is used to process real-time acquired nonlinear structural monitoring data to determine whether the sensor system has faults and which sensors have failed.

2. The sensor fault detection method for a structural health monitoring system as described in claim 1, characterized in that, The time index of the mixture t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points within the auxiliary reference plane during the same sampling time. Using the γ quadrant coordinates of these two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane, including: Randomly determine a reference surface and a surface to be moved. Determine the coordinates of any two points and the coordinates of the first midpoint of any two points on the reference surface, where the reference surface is any surface. Time index of the mixture t Determine the coordinates of two accompanying points on the surface to be moved and the coordinates of the second midpoint of the line connecting the two accompanying points for any two points in the reference plane within the same sampling time. Using a preset first formula to process the coordinates of any two points, the coordinates of two accompanying points, and the coordinates of the first midpoint, the first γ quadrant coordinate value of the accompanying point coordinates of the first midpoint on the surface to be moved is obtained. Using a preset second formula to process the first γ quadrant coordinate value and the second midpoint coordinate, the second γ quadrant coordinate value of the accompanying point of the first midpoint after the move is obtained. The expression (1) of the first preset formula is: in, This represents the γ quadrant coordinate value of the first midpoint between any two points on the reference plane. This represents the γ quadrant coordinate value of one of any two points on the reference plane; This represents the γ quadrant coordinate value of the other point among any two points on the reference plane. This represents the γ quadrant coordinate value of one of the two accompanying points on the surface to be moved; This represents the γ quadrant coordinate value of the other point among the two accompanying points on the surface to be moved. The expression (2) of the second preset formula is: in, Points on the surface to be moved p c The γ quadrant coordinates before movement; Points on the surface to be moved p c The shifted γ quadrant coordinates For learning rate, The range of values ​​is 0 < <1; The γ quadrant coordinates of the accompanying point of the first midpoint coordinates after and before the move are processed using the third preset formula to obtain the error value. If the error value is greater than the preset first threshold, the γ quadrant coordinates of the accompanying point of the first midpoint coordinates after the move are iteratively calculated. The expression (3) of the three preset formulas is: In the formula N k This represents the number of points to be updated in each iteration, and n represents the number of observed signals; The γ quadrant coordinates of the accompanying point represent the coordinates of the first midpoint after the movement. The γ quadrant coordinates of the accompanying point represent the coordinates of the first midpoint before the movement.

3. The sensor fault detection method for a structural health monitoring system as described in claim 2, characterized in that, If the error value is less than the preset first threshold, the iteration ends, and the γ quadrant coordinate values ​​of all the moved first midpoint coordinates generated by the iteration are sorted in ascending order and assigned to each signal to be separated. The fourth preset formula is used to perform a smooth fit on each signal to be separated to obtain the signal function to be separated; The expression (4) of the fourth preset formula is: z i s The signal to be separated L The time window length is defined as a value within the range of (0, 1). N () is an integer, where N is the number of samples; The FastICA model is used to process the signal function to be separated, and multiple independent elements representing the independent signals are obtained. The separation matrix is ​​then determined based on these independent elements.

4. The sensor fault detection method for a structural health monitoring system as described in claim 3, characterized in that, The step of determining the separation matrix based on multiple independent elements includes: By using the Euclidean norm to process all independent elements, we obtain a sequence of independent elements; By using the top-ranked independent elements as the dominant independent elements, a separation matrix is ​​obtained.

5. The sensor fault detection method for a structural health monitoring system as described in claim 4, characterized in that, The process of using an improved geometric post-nonlinear independent element analysis model to process real-time acquired nonlinear structural monitoring data and determine whether a fault exists in the sensor system includes: The Euclidean norm is used to process the separation matrix to obtain the separation principal part matrix and the cofactor matrix; The independent element vectors of the principal part and the independent element vector of the remainder are calculated by multiplying the principal part and the remainder of the separation matrix with the nonlinear data acquired at any time. The independent element vectors of the main part, the independent element vectors of the remainder, and the monitoring data during the operation of the sensor system are processed according to the fifth preset formula to obtain the normal monitoring statistics and real-time monitoring statistics. The kernel density estimation algorithm is used to process the normal monitoring statistics to obtain the normal monitoring statistics threshold. The fifth preset formula includes I 2 Expression for monitoring statistics (5) I e 2 The expression for the monitoring statistics (6) and SPE The expression for the monitoring statistics (7); In the formula, , , Represents the principal matrix. This represents the monitoring data during system operation. Represents the remainder matrix. Denotes the independent element vectors of the principal component. Denotes the independent element vectors of the remainder. ,in, This is the whitening matrix; If the real-time monitoring statistic value is greater than the normal monitoring statistic threshold, then the sensor system is determined to be faulty; otherwise, the sensor system is not faulty.

6. The sensor fault detection method for a structural health monitoring system as described in claim 5, characterized in that, The process of using an improved geometric post-nonlinear independent element analysis model to process real-time acquired nonlinear structure monitoring data and identify faulty sensors in the sensor system includes: Contribution analysis algorithms are used to process real-time monitoring statistics. The contribution analysis formula includes: I 2 The expression for the contribution value (8) I e 2 The expression for the contribution value (9) SPE The expression for the contribution value (10): In the formula, express I 2 The contribution value; express I e 2 The contribution value; express SPE The contribution value.

7. The sensor fault detection method for a structural health monitoring system as described in claim 6, characterized in that, After identifying the faulty sensor in the sensor system, the following steps are also included: Sensor fault types are determined using a mathematical model of sensor faults. These fault types include: fixed deviation fault, linear deviation fault, constant gain fault, accuracy degradation fault, jamming fault, white noise jamming fault, and zero line drift fault.

8. A sensor fault detection device for a structural health monitoring system, characterized in that, include: The data acquisition module is used to acquire nonlinear structural monitoring data collected by the sensor system of the structural health monitoring system; The model determination module is used to process nonlinear structural monitoring data using an improved geometric PNL mixture model to obtain a linear mixture signal to be separated. The FastICA model is then used to process the mixture signal to be separated, obtaining multiple independent elements. Based on these independent elements, the separation matrix of the improved geometric post-nonlinear independent element analysis model is determined. In the improved geometric PNL mixture model, the time exponent of the mixture is used to determine the separation matrix. t The γ quadrant coordinates are used to determine the γ quadrant coordinates of two accompanying points on the surface to be moved for any two points in the auxiliary reference plane within the same sampling time. Using the γ quadrant coordinates of the two accompanying points on the surface to be moved, the γ quadrant coordinates of all points on the surface to be measured are calculated when the points are moved to the auxiliary reference plane. The fault determination module is used to process real-time acquired nonlinear structural monitoring data using an improved geometric post-nonlinear independent element analysis model to determine whether the sensor system has a fault and which sensor has a fault.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the sensor fault detection method of the structural health monitoring system as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or instruction set is loaded and executed by a processor to implement the sensor fault detection method of the structural health monitoring system as described in any one of claims 1-7.

Images