A structural strain monitoring method based on time series relative difference

Abstract

The application discloses a structure strain monitoring method based on time series relative difference, which comprises the following steps: collecting original strain time series data; adopting a statistical method based on a sliding window, identifying and processing abnormal data points in the original strain time series data to obtain purified strain data; adopting a time difference technique, calculating the difference value of strain monitoring data separated by a fixed time interval to obtain relative strain reflecting load change; combining field deformation test data to generate a load strain-deformation coupling curve, and evaluating the structure health state by analyzing the long-term evolution form change of the curve. The application organically combines the moving Z-score method with the time difference technique, has a synergistic effect, purifies the original strain data from abnormal points, can prevent extreme abnormal values from interfering with subsequent difference calculation, and forms a closed-loop solution from original data collection to quality analysis curve generation.

Description

Technical Field

[0001] This invention relates to the field of strain measurement, and in particular to a structural strain monitoring method based on time series relative difference. Background Technology

[0002] Strain measurement is a crucial process for assessing the behavior of materials and structures under stress, providing essential data for evaluating structural integrity and performance. Its applications are wide-ranging and vital, encompassing civil infrastructure (such as bridges, dams, and buildings), aerospace (such as aircraft frames), and renewable energy systems (such as wind turbines). The goal of structural health monitoring is to build a system similar to the human nervous system, capable of sensing structural "damage" or "lesions," thereby providing a basis for subsequent maintenance decisions. Strain data is highly valuable because it is highly sensitive to localized damage, revealing early signs before problems such as cracks or material fatigue become severe.

[0003] In any application requiring long-term continuous strain monitoring, whether the object is a bridge, a wind turbine, or other large structure, the quality of monitoring data generally faces three core challenges: temperature drift, zero-point drift, and data anomalies. These challenges stem from environmental factors, sensor characteristics, and external interference, collectively constituting the main obstacles to long-term monitoring data processing.

[0004] Temperature drift is one of the most challenging aspects of long-term monitoring. It is defined as any change in strain gauge resistance not caused by structural mechanical stress. Its physical mechanism is quite complex, stemming from the superposition of two effects: first, the resistivity of the strain gauge grid wire alloy itself changes with temperature; second, due to the coefficient of thermal expansion of the strain gauge material… Typically related to the coefficient of thermal expansion of the material being tested. Unlike other components, temperature changes cause relative expansion or contraction between the two. This mismatched deformation exerts a mechanical stretching or compression on the strain gauge, resulting in a "thermal output" or "apparent strain" that is not caused by external loads. In scenarios requiring long-term monitoring across seasons (e.g., from summer to winter), large temperature differences can cause significant drift, the magnitude of which may even completely mask the true structural strain signal, leading to data failure.

[0005] Zero-point drift refers to the slow, gradual shift of the sensor output baseline over time, even under constant temperature and no external load conditions. This phenomenon is primarily attributed to the long-term aging and creep effects of the sensor materials (such as the sensing grid and substrate) and the adhesives used to fix them to the structure. This drift accumulates with prolonged service life, severely impacting the absolute accuracy of measurements and the consistency of long-term data.

[0006] Outliers refer to data points that deviate significantly from other observations in a data sample. In raw data streams of strain monitoring, such abnormal data frequently occur, with various sources.

[0007] To address the above three core challenges, the existing technology has developed various methods, including:

[0008] Hardware compensation scheme: For temperature drift, the most classic method is to introduce a "compensation piece" in the bridge. The compensation piece is installed on a material of the same material as the measured structure but completely free of force, and is placed in the same temperature environment. In theory, the temperature drift it produces can offset the temperature drift of the working strain piece. However, this method significantly increases the complexity, cost and potential failure points of the system. It requires the temperature field of the compensation piece to be completely consistent with that of the working piece, which is extremely difficult to guarantee in practical applications of large structures.

[0009] Software and auxiliary sensor scheme: Another common method is to install an additional temperature sensor to measure the ambient temperature in real time, and then compensate according to the pre-calibrated "temperature-strain" relationship curve through a software algorithm. This method also requires additional hardware, increasing the maintenance burden, and the universality and long-term effectiveness of the calibration curve are questionable. For zero drift, the conventional approach is to periodically calibrate the sensor offline, but this is obviously impractical for wind turbines or important bridges that need to operate continuously.

[0010] Traditional abnormal data processing algorithm: For outliers, traditional methods rely on simple statistical threshold methods (such as setting fixed upper and lower limits) or basic filtering algorithms (such as median filtering, Kalman filtering, etc.). These methods have limited effectiveness in handling complex and non-stationary signals, are difficult to adapt to changing abnormal patterns, and lack the processing capacity for large-scale real-time data streams. Although there are advanced deep learning methods such as long short-term memory networks (LSTM), their high computational cost, dependence on a large amount of labeled training data, and model complexity limit their widespread application on resource-constrained structural health monitoring edge devices.

[0011] In summary, the current technical field lacks a comprehensive software solution that can simultaneously, efficiently and cost-effectively address the three core problems of temperature drift, zero drift and data anomalies. In particular, existing methods often strive to "correct" or "compensate" to obtain the "true value" of absolute strain, which itself faces significant challenges. As infrastructure service life continues to extend, the market urgently needs a more intelligent, robust and economical data processing method to improve the accuracy and reliability of long-term structural health monitoring. SUMMARY

[0012] In order to solve the above technical problems, the application provides a structural strain monitoring method based on time series relative difference.

[0013] The technical scheme for solving the above technical problems is: a structural strain monitoring method based on time series relative difference, comprising the following steps:

[0014] S1: collecting original strain time series data from strain sensors installed on the structure;

[0015] S2: using a statistical method based on a sliding window to identify and process abnormal data points in the original strain time series data, to obtain purified strain data;

[0016] S3: for the purified strain data, using time difference technology, by calculating the difference of strain monitoring data separated by a fixed time interval, to obtain relative strain reflecting load change;

[0017] S4: using the relative strain reflecting load change, combined with field deformation test data, to generate load strain-deformation coupling curve, by analyzing the long-term evolution form change of the load strain-deformation coupling curve, to evaluate the health status of the structure.

[0018] The above structural strain monitoring method based on time series relative difference, in the step S2, the statistical method based on the sliding window is the moving Z-score, the moving Z-score identifies abnormal points by calculating the relationship between the data point and its local mean and local standard deviation in the sliding window.

[0019] The above structural strain monitoring method based on time series relative difference, in the step S2, the moving Z-score is an online anomaly detection algorithm suitable for time series, which evaluates the local abnormality degree of the data point through a sliding window; for the data point , The moving Z-score value is defined as:

[0020]

[0021] Wherein, and are the mean and standard deviation of the W data points before , and W is the window size;

[0022] Set an abnormal threshold T, if the moving Z-score value of a data point is greater than T, the data point is marked as abnormal; for the identified abnormal data, use the preceding segment replacement strategy for repair.

[0023] The structural strain monitoring method based on time series relative difference, in the step S2, the pre-segment replacement strategy is specifically:

[0024] For a single abnormal data point: directly replace the value of the abnormal data point with the value of its previous normal data point;

[0025] For continuous abnormal segments: if an abnormal segment composed of k consecutive data points is detected, locate the normal data segment with the same length k adjacent to the abnormal segment, and then replace the entire abnormal segment with this normal data segment.

[0026] The structural strain monitoring method based on time series relative difference, in the step S3, the calculation formula of the measured strain value of the purified time point

[0027]

[0028] Among them, is the true mechanical strain caused by structural stress; is the apparent strain caused by temperature change; is the zero drift caused by sensor aging factors.

[0029] The structural strain monitoring method based on time series relative difference, in the step S3, the process of obtaining the relative strain reflecting load change is:

[0030] Select a fixed time interval , for , backtracking to find the historical purified data point before time, and subtracting operation to get the relative strain reflecting load change :

[0031]

[0032] It can be obtained:

[0033]

[0034] Based on the characteristics of the slow change of the drift signal, when is selected to be short enough, it is assumed that:

[0035]

[0036]

[0037] Therefore, the drift term is effectively canceled out in the subtraction process, and the formula is simplified as: ​​​

[0038] .

[0039] The structural strain monitoring method based on time series relative difference, in the step S3, The value of the characteristic time is greater than the characteristic time of the structure in a typical working condition for one time, and is less than the thermal response time constant of the strain sensor caused by the temperature change of the environment.

[0040] The structural strain monitoring method based on time series relative difference, in the step S4, the long-term evolution form of the load strain-deformation coupling curve is continuously monitored to determine whether the structural stiffness decreases, thereby realizing early warning and quantitative evaluation of structural damage.

[0041] The beneficial effects of the present application are:

[0042] 1. For the instantaneous abnormal data caused by human touch, circuit failure and other reasons, the present application first uses the moving Z-score algorithm to identify and eliminate the original data stream. This algorithm calculates the deviation of the current data point relative to the local statistical characteristics (mean and standard deviation) in the window by sliding a window on the data stream. Compared with the traditional global threshold method, the moving Z-score can adapt to the local dynamic changes of the data, and has better robustness for non-stationary time series. Compared with deep learning models, it is extremely efficient in calculation and does not require a training process, which is very suitable for real-time processing of large-scale data at the data acquisition front end, significantly improving data quality and processing efficiency.

[0043] 2. After the abnormal points are processed, the present application calculates the difference between the monitoring data separated by a fixed time to obtain the relative strain reflecting the load change. Since temperature drift and zero drift are slow-changing processes, their drift amounts are approximately constant in a short time interval, so they are effectively canceled out in the subtraction operation. This method does not require any additional hardware devices such as temperature sensors or compensation sheets, fundamentally simplifying the monitoring system, reducing the initial cost, wiring complexity and long-term maintenance cost, and converting a complex physical compensation problem into a simple mathematical calculation problem, which is especially suitable for long-term monitoring environments with large temperature differences and can simultaneously eliminate two main slow drifts.

[0044] 3、The application combines the moving Z-score method with the time difference technique, forming a logical and efficient data purification process of "first removing outliers, then eliminating drifts". This process design has a synergistic effect. First, the original data is purified to prevent extreme outliers from interfering with subsequent difference calculations, so that more stable and reliable relative values are obtained. The entire process forms a closed-loop solution from raw data collection to generating high-quality analysis curves, and is particularly suitable for applications with extremely high reliability requirements for long-term continuous monitoring data.

[0045] 4、The application specifies the selection principle of the fixed time interval , that is It must meet the dual constraints: it must be greater than the characteristic time of a complete working condition (such as fan variable pitch, vehicle passing bridge) of the structure, so as to capture the complete mechanical response process; at the same time, it must be much smaller than the thermal response time constant of the structure, so as to ensure that the drift is approximately constant within Δt. This mechanism directly links the algorithm parameters to the physical characteristics (working condition period, thermodynamic characteristics) of the monitored object, providing theoretical support for the effectiveness of the method and making it portable and configurable, which can be optimized and adjusted according to different monitoring objects and environments. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is the overall flowchart of the application.

[0047] Figure 2 is the principle diagram of moving Z-score.

[0048] Figure 3 is the waveform diagram before the abnormal data points are corrected.

[0049] Figure 4 is the waveform diagram after the abnormal data points are corrected.

[0050] Figure 5 is the strain coupling diagram before processing using the time difference technique.

[0051] Figure 6 is the strain coupling diagram after processing using the time difference technique.

[0052] Figure 7 is the structure damage evolution process curve. DETAILED DESCRIPTION

[0053] The application will be further described below in conjunction with the drawings and examples.

[0054] As Figure 1 shown, a structure strain monitoring method based on time series relative difference includes the following steps:

[0055] S1: Acquire raw strain time series data from strain sensors mounted on the structure.

[0056] S2: A sliding window-based statistical method is used to identify and process outlier data points in the original strain time series data to obtain purified strain data.

[0057] like Figure 2 As shown, the statistical method based on the sliding window is the moving Z-score. The moving Z-score identifies outliers by calculating the relationship between data points and their local means and local standard deviations within the sliding window. The moving Z-score is an online anomaly detection algorithm suitable for time series data. It uses a sliding window to evaluate the degree of local anomalies in data points; for data points in a time series... , Z-score value of movement Defined as:

[0058]

[0059] in, and They are The mean and standard deviation of the previous W data points; the moving Z-score visually represents the standard deviation by which the current data point deviates from its "recent historical" mean.

[0060] W represents the window size, a key parameter that determines the locality of the statistical baseline. Smaller windows are sensitive to short-term fluctuations, while larger windows are smoother but may miss transient anomalies. The window size should be determined based on the sampling frequency to cover a representative time period.

[0061] An anomaly threshold T is set, which is based on the "3-sigma" principle of normal distribution, that is, about 99.7% of the data should fall within ±3 standard deviations of the mean; if the moving Z-score of a data point is greater than T, the data point is marked as an anomaly; for the identified anomalous data, a preceding segment replacement strategy is used for repair.

[0062] The specific strategy for replacing preceding segments is as follows:

[0063] For a single outlier data point: directly replace the value of the outlier data point with the value of the preceding normal data point;

[0064] For consecutive abnormal segments: If an abnormal segment consisting of k consecutive data points is detected, the preceding normal data segment of the same length k is located, and then the entire abnormal segment is replaced with this normal data segment.

[0065] In structural monitoring data with high sampling rate, the value change between adjacent data points is continuous and small. Therefore, using the value of the immediately preceding normal data point to replace the abnormal data point is based on the reasonable assumption that "if no abnormality occurs, the signal value should remain smooth". This method can maximize the local dynamic characteristics and time continuity of the data, avoiding the introduction of artificial mutations, which is crucial for subsequent differential calculations that rely on data time correlation. This method only needs to backtrack one or a few data points, with minimal computational effort, fully meeting the requirements of real-time processing. At the same time, it strictly adheres to the causality of time series, using only historical data for correction, avoiding the potential forward-looking bias that may be brought by using future data (such as the next point in linear interpolation).

[0066] By immediately correcting the abnormal point, it can prevent the abnormal value from entering the sliding window of the subsequent moving Z-score calculation, thereby avoiding its pollution of the mean and standard deviation within the window, and ensuring the accuracy of subsequent anomaly detection.

[0067] As shown in Figure 3 , there is a clear spike anomaly in the original data. By using the pre-segment replacement, the abnormal data point is replaced by the value of its previous normal data point, and the corrected data waveform is shown in Figure 4 , which restores the local smoothness and continuity.

[0068] S3: For the purified strain data, use time difference technique to calculate the difference of strain monitoring data with a fixed time interval, to obtain the relative strain reflecting load change.

[0069] The measured strain value of the purified time is calculated by the formula:

[0070]

[0071] Where, is the true mechanical strain caused by structural stress; is the apparent strain caused by temperature change; is the zero drift caused by sensor aging factors.

[0072] The process of obtaining the relative strain reflecting load change is:

[0073] Select a fixed time interval , for , backtrack to find the historical purified data point before time , and perform subtraction operation to obtain the relative strain reflecting load change : ​

[0074]

[0075] Available:

[0076]

[0077] Based on the characteristics of the drift signal's slow variation, when is chosen to be short enough, it is assumed that:

[0078]

[0079]

[0080] Therefore, the drift term is effectively canceled out in the subtraction process, and the formula is simplified to:

[0081] .

[0082] In this way, the corresponding strain mainly reflects the real mechanical response changes of the structure in time period, and is basically not disturbed by slow drift factors, such as Figure 5 , Figure 6 .

[0083] The choice of is the key of the present application, which directly affects the effect of drift elimination and the integrity of mechanical information capture. The value of must be balanced between two mutually restrictive conditions:

[0084] Lower limit constraint: greater than the working condition period, that is, greater than the characteristic time of a typical working condition of the structure in operation. It must be long enough to include one or more complete typical working condition cycles of the monitored structure. For example, for a wind turbine, should include a complete cycle of significant load changes such as yaw, variable pitch, or wind speed fluctuations. If is too short, the difference operation may only capture a partial segment of the working condition, resulting in relative values that cannot reflect the complete mechanical behavior.

[0085] Upper limit constraint: less than the temperature drift significant change time, that is, less than the thermal response time constant of the strain sensor caused by significant temperature drift due to environmental temperature changes. It must be short enough to ensure that the apparent strain increment caused by temperature changes is negligible within this time period. Large structures usually have large thermal masses and long thermal time constants, and their temperature response is relatively slow. Experience and experimental data show that environmental temperature changes are usually not enough to cause significant drift of the strain gauge on a 15-minute time scale.

[0086] Handling extreme operating conditions: Taking wind turbine monitoring as an example, when events such as emergency shutdowns or extreme gusts occur, the duration may exceed [a certain period of time]. In extreme operating conditions, a single difference may not capture the full picture of the event. However, this does not affect the long-term effectiveness of this method. Because if extreme conditions cause damage to the structure (such as plastic deformation or crack propagation), it will inevitably lead to permanent changes in the structure's fundamental physical properties, such as stiffness. This change will be continuously reflected in the load-strain-deformation coupling curves of all subsequent normal operating conditions; for example, the slope of the curve will permanently change. Therefore, by monitoring the evolution of this curve shape over a long period, the cumulative impact of extreme conditions on structural health can still be effectively identified.

[0087] S4: By utilizing relative strain, which reflects load changes, and combining it with on-site deformation test data, a load-strain-deformation coupling curve is generated. Through analysis of the long-term evolution of this curve (such as slope, linearity, and hysteresis loop area), early warning and quantitative assessment of structural damage can be achieved, thus evaluating the structural health status. Figure 7 As shown, Figure 7 The typical load-strain-deformation coupling curves of the structure at three different service time points (10 days, 6 months, and 1 year) are shown. Figure 7 As can be seen, the curve's shape undergoes a significant evolution with prolonged service life: on the one hand, the overall slope of the curve gradually decreases, directly reflecting the degradation of structural stiffness; on the other hand, the area enclosed by the hysteresis loop formed by the curve increases significantly over time, representing increased energy dissipation due to damage mechanisms such as microcrack propagation during load cycles. Therefore, long-term monitoring of this curve's shape evolution allows for a clear tracking of the structure's progression from an initial healthy state (10 days) to gradual damage accumulation (one year), thereby enabling a quantitative assessment of the structure's health status.

Claims

1. A structural strain monitoring method based on time series relative difference, characterized in that, Includes the following steps: S1: Acquire raw strain time series data from strain sensors mounted on the structure; S2: A sliding window-based statistical method is used to identify and process outlier data points in the original strain time series data to obtain purified strain data; S3: For the strain data after purification, time difference technology is used to obtain the relative strain reflecting the load change by calculating the difference between strain monitoring data with a fixed time interval between them. Time interval The value of is greater than the characteristic time for the structure to complete one cycle under typical operating conditions, and less than the thermal response time constant caused by significant temperature drift of the strain sensor due to changes in ambient temperature. S4: By using relative strain, which reflects load changes, and combining it with on-site deformation test data, a load-strain-deformation coupling curve is generated. By analyzing the long-term evolution of the load-strain-deformation coupling curve, the health status of the structure is assessed.

2. The structural strain monitoring method based on time series relative difference according to claim 1, characterized in that, In step S2, the statistical method based on the sliding window is the moving Z-score. The moving Z-score identifies outliers by calculating the relationship between data points and their local mean and local standard deviation within the sliding window.

3. The structural strain monitoring method based on time series relative difference according to claim 2, characterized in that, In step S2, the moving Z-score is an online anomaly detection algorithm suitable for time series data. It uses a sliding window to evaluate the degree of local anomalies in data points. , Z-score value of movement Defined as: ； in, and They are The mean and standard deviation of the previous W data points, where W is the window size; Set an anomaly threshold T. If the moving Z-score of a data point is greater than T, the data point is marked as an anomaly. For the identified anomaly data, a preceding segment replacement strategy is used for repair.

4. The structural strain monitoring method based on time series relative difference according to claim 3, characterized in that, In step S2, the preceding segment replacement strategy is specifically as follows: For a single outlier data point: directly replace the value of the outlier data point with the value of the preceding normal data point; For consecutive abnormal segments: If an abnormal segment consisting of k consecutive data points is detected, the preceding normal data segment of the same length k is located, and then the entire abnormal segment is replaced with this normal data segment.

5. The structural strain monitoring method based on time series relative difference according to claim 1, characterized in that, In step S3, the time after purification Measured strain value The calculation formula is: ； in, It is the actual mechanical strain caused by structural stress; It is the apparent strain caused by temperature change; The zero-point drift is caused by sensor aging.

6. The structural strain monitoring method based on time series relative difference according to claim 5, characterized in that, In step S3, the process of obtaining the relative strain reflecting the load change is as follows: Select a fixed time interval ,for Backtracking search Historical cleanup data points before the specified time Then, a subtraction operation is performed to obtain the relative strain reflecting the load change. : ； We can obtain: ； Based on the slowly varying characteristics of drift signals, when When the selection is short enough, assume: ； ； Therefore, the drift term is effectively canceled out during the subtraction process, and the formula simplifies to: 。 7. The structural strain monitoring method based on time series relative difference according to claim 1, characterized in that, In step S4, by continuously monitoring whether the long-term evolution of the load-strain-deformation coupling curve changes, it can be determined whether the structural stiffness has decreased, thereby achieving early warning and quantitative assessment of structural damage.

Images