A target machine shooting intelligent monitoring method and system based on big data

Abstract

The present application relates to target equipment fault diagnosis technical field, disclose a kind of based on big data's target machine shooting intelligent monitoring method and system, the present application is hit event as time reference synchronous acquisition transmission horizontal axis transient impact vibration signal and rise and fall motor current waveform, false mutation point is eliminated by adaptive filtering, with impact energy peak value alignment intercept waveform fragment, extract peak amplitude, energy integral and duration from impact waveform, extract current rising slope, rise and fall period and current integral from current waveform, splice as individualized sample feature vector after standardization, utilize the same target machine historical data without exception to build individualized normal model including single dimension distribution form parameter and dimension intercovariance coefficient matrix, in subsequent shooting, current feature vector is input model, and each characteristic deviation score and Mahalanobis distance are calculated, realize the comprehensive determination of each dimension deviation degree and multi-dimensional joint probability abnormal degree.

Description

Technical Field

[0001] This invention relates to the field of target drone equipment fault diagnosis technology, and more specifically, to a target drone intelligent monitoring method and system based on big data. Background Technology

[0002] Existing technologies have made some progress in terms of hit reporting accuracy and mechanical structure reliability. For example, patent CN212133460U discloses a hit reporting device based on shock wave and vibration combined detection, which improves the reliability of shooting event discrimination by combining vibration and shock wave detection modules. Patent CN215296022U discloses a target drone with built-in sensors, which installs an acceleration sensor on the transmission horizontal shaft to enhance the stability of signal acquisition. Patent CN116520791B discloses a fully automatic detection method for target drone safety controllers based on signal threshold technology, which carries out fully automated functional testing of safety controller components.

[0003] However, considering the aforementioned technological achievements and the actual status of target drone operation and maintenance, the focus is mainly on the accuracy of hit reporting for a single shot and the electrical performance testing of individual components. There is a lack of systematic quantitative analysis and joint diagnostic methods regarding the correlation between the transient impact vibration signal transmitted through the transmission shaft during target firing and the synchronously generated electromechanical response waveform of the tilting motor. In practice, when the target drone's mechanical condition slowly deteriorates due to bearing jamming, increased transmission clearance, or reducer wear, the energy characteristics of the vibration impact and the rising slope of the motor drive current often deviate synchronously. Its historical distribution is normal, but on-site maintenance personnel can only perceive this degradation process through passive manual inquiry or daily maintenance experience. Therefore, how to use each hit event as a time benchmark to synchronously collect and perform feature correlation analysis on the transient impact vibration signal of the transmission horizontal shaft and the current waveform of the starting and stopping motor, and construct a unique individual normal model for each target drone, so as to realize the electromechanical collaborative health status assessment of the target drone based on big data, has become a difficult problem in the current intelligent equipment support of live-fire ranges. In order to solve this technical problem, we provide a target drone intelligent monitoring method and system based on big data. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent monitoring of target drone shooting based on big data, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, one of the objectives of this invention is to provide a big data-based intelligent monitoring method for target drone firing, comprising the following steps: S1. During the operation of the target machine, the transient impact vibration signal transmitted through the transmission horizontal shaft after the target plate is impacted is synchronously collected, and the current waveform of the tilting motor during the driving cycle of the hit event is collected, taking the moment of the hit event as the time reference, to form an impact vibration sequence and current waveform sequence uniquely associated with each hit event. S2. Apply adaptive filtering to the transient impact vibration signal to eliminate false abrupt change points, and use the peak time of the impact energy in the transient impact vibration signal as the alignment reference to extract the impact waveform segment containing the main waveform of the impact response. At the same time, extract the current waveform segment from the current waveform of the starting and stopping motor that is time-aligned with the impact waveform segment. S3. Extract the peak amplitude of the impact response, the integral amount of the impact energy, and the duration of the impact response from the impact waveform segment to form an impact feature vector. Extract the current rise slope, the start-up / reverse cycle, and the integral amount of the current from the current waveform segment to form an electromechanical response feature vector. Then, concatenate the impact feature vector and the electromechanical response feature vector to form an individualized sample feature vector. S4. Using big data technology, we use the individualized sample feature vectors corresponding to all historical hit events of the same target machine under no mechanical abnormality as training data to construct a unique individualized normal model of the target machine. S5. In subsequent target practice missions, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension and the degree of abnormality of the multi-dimensional joint probability.

[0006] The second objective of this invention is to provide a system for implementing a big data-based intelligent monitoring method for target drone shooting, comprising: The signal synchronization acquisition unit is used to synchronously acquire the transient impact vibration signal transmitted through the transmission horizontal shaft after the target plate is impacted, as well as the current waveform of the tilting motor during the driving cycle of the hit event, with the time of the hit event as the time reference, during the operation of the target machine, to form an impact vibration sequence and current waveform sequence that are uniquely associated with each hit event. The signal processing and segment extraction unit is used to apply adaptive filtering to the transient impact vibration signal to eliminate false abrupt change points, and to extract an impact waveform segment containing the main waveform of the impact response based on the peak time of the impact energy in the transient impact vibration signal. At the same time, it extracts a current waveform segment from the current waveform of the starting and stopping motor that is time-aligned with the impact waveform segment. The feature extraction and standardization unit is used to extract the peak amplitude of the impact response, the integral amount of the impact energy, and the duration of the impact response from the impact waveform segment to form an impact feature vector, and to extract the current rise slope, the start-up and reversal period, and the integral amount of the current from the current waveform segment to form an electromechanical response feature vector. The impact feature vector and the electromechanical response feature vector are then standardized and concatenated to form an individualized sample feature vector. The model building and anomaly judgment unit is used to construct a unique individualized normal model for the target machine using the individualized sample feature vectors corresponding to all historical hit events of the same target machine under no mechanical abnormality conditions. In subsequent target shooting tasks, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension and the degree of multi-dimensional joint probability anomaly.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention synchronously acquires the transient impact vibration signal of the transmission horizontal shaft and the current waveform of the starting and stopping motor using the hit event as the time reference. Waveform segments are extracted using the peak moment of impact energy as the alignment reference. Impact features and electromechanical features are extracted from the two types of signals, respectively. After standardization, they are spliced ​​into an individualized sample feature vector. Based on this, an individualized normal model containing one-dimensional distribution morphological parameters and inter-dimensional covariance coefficient matrices is constructed using historical anomaly-free data of the same target machine. In subsequent target firing, the degree of anomaly of the current hit event is comprehensively evaluated by one-dimensional deviation score and multi-dimensional Mahalanobis distance. This effectively solves the problem that existing target machine monitoring methods rely only on a single vibration signal, ignore electromechanical coupling degradation correlation, and cannot accurately model the individual differences of each target machine. It achieves the effect of early identification of mechanical state anomalies from the electromechanical coordinated response of a single hit event and improves the intelligent level of target machine health management. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the overall workflow of the present invention; Figure 2 This is a schematic diagram of the overall structure of the present invention; Detailed Implementation

[0009] 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.

[0010] Please see Figure 1As shown, one of the objectives of this embodiment is to provide a big data-based intelligent monitoring method for target drone shooting, including the following steps: S1. During the operation of the target machine, the transient impact vibration signal transmitted through the transmission horizontal shaft after the target plate is impacted is synchronously collected, and the current waveform of the tilting motor during the driving cycle of the hit event is collected, taking the moment of the hit event as the time reference, to form an impact vibration sequence and current waveform sequence uniquely associated with each hit event. S2. Apply adaptive filtering to the transient impact vibration signal to eliminate false abrupt change points, and use the peak time of the impact energy in the transient impact vibration signal as the alignment reference to extract the impact waveform segment containing the main waveform of the impact response. At the same time, extract the current waveform segment that is time-aligned with the impact waveform segment from the current waveform of the starting and stopping motor. S3. Extract the peak amplitude of the impact response, the integral amount of the impact energy, and the duration of the impact response from the impact waveform segment to form an impact feature vector. Extract the current rise slope, the start-up and reversal period, and the integral amount of the current from the current waveform segment to form an electromechanical response feature vector. Then, concatenate the impact feature vector and the electromechanical response feature vector to form an individualized sample feature vector. S4. Using big data technology, we use the individualized sample feature vectors corresponding to all historical hit events of the same target machine under no mechanical abnormality as training data to construct a unique individualized normal model of the target machine. S5. In subsequent target practice missions, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension and the degree of abnormality of the multi-dimensional joint probability.

[0011] S2 applies adaptive filtering to the transient impact vibration signal, specifically including: After the transient impact vibration signal corresponding to each hit event is collected, the ratio of the local energy of the transient impact vibration signal in the preset first time period to the global average energy in the preset second time period is calculated. When the ratio exceeds the dynamic threshold obtained by statistical analysis of the target machine's historical data on no abnormal hit events, false abrupt changes are identified within the preset first time period. The signal amplitude of the transient impact vibration signal in the predetermined area of ​​the false abrupt change point is replaced by the signal amplitude of the effective impact waveform segments before and after it to eliminate the false abrupt change point.

[0012] S2 uses the peak moment of the impact energy in the transient impact vibration signal as the alignment reference, specifically including: On the transient impact vibration signal after the elimination of false abrupt change points, the square of the signal amplitude is calculated and integrated over time to obtain the impact energy accumulation curve. The inflection point on the impact energy accumulation curve where the slope changes from a continuous rapid increase to a gradual change is identified, and the time corresponding to this inflection point is determined as the peak time of the impact energy of the main waveform of the impact response.

[0013] S3 extracts the peak amplitude of the impact response, the integral of the impact energy, and the duration of the impact response from the impact waveform segment to form the impact feature vector, which specifically includes: Centered on the peak moment of the impact energy, a fixed-length window is extended to both sides of the time axis. The signal within the window is defined as an impact waveform segment. Within the impact waveform segment, the maximum absolute value of the signal is taken as the peak amplitude of the impact response. The result of integrating the square of the signal amplitude over the time range of the impact waveform segment is taken as the impact energy integral. The time from when the signal amplitude first exceeds the noise threshold to when it finally falls back below the noise threshold within the impact waveform segment is taken as the impact response duration. The peak amplitude of the impact response, the impact energy integral, and the impact response duration are arranged in order to form the impact feature vector.

[0014] S3 extracts the current rise slope, start-up / reversal period, and current integral from the current waveform segment to form the electromechanical response feature vector, which specifically includes: Within the current waveform segment, the starting point of each current rise from zero and the ending point of the current fall back to zero are identified. The time length between one starting point and the next starting point is defined as a complete start-down drive cycle. Within a single start-down drive cycle, the data points of the current rise phase are fitted with a straight line. The slope of the resulting straight line is taken as the current rise slope of a single start-down drive cycle. The time length of a single start-down drive cycle is taken as the start-down cycle. The result of integrating the absolute value of the current within the time range of a single start-down drive cycle is taken as the current integral. From the multiple start-down drive cycles contained in the current waveform segment, the start-down drive cycle with the highest temporal overlap with the impact waveform segment is selected. The current rise slope, start-down cycle, and current integral extracted within the highest start-down drive cycle are arranged in order to form an electromechanical response feature vector.

[0015] In S3, the impact feature vector and the electromechanical response feature vector are concatenated to form an individualized sample feature vector, specifically including: Before constructing a unique individualized normal model for the target drone, the historical average and standard deviation of each characteristic component in the impact characteristic vector and the historical average and standard deviation of each characteristic component in the electromechanical response characteristic vector are calculated using all historical hit event data of the target drone under no mechanical abnormality conditions. For any hit event, after obtaining the impact feature vector and the electromechanical response feature vector, the value of each feature component is subtracted from the historical average value corresponding to the feature component, and then divided by the historical standard deviation corresponding to the feature component to complete the standardization process. The standardized impact feature vector and the electromechanical response feature vector are then connected in sequence to form an individualized sample feature vector.

[0016] In S4, a unique, individualized normal model specific to the target machine is constructed, including: The individualized sample feature vectors corresponding to all historical events without mechanical abnormality are standardized and then combined to form an individualized sample feature vector set, which is regarded as the feature sample group under the normal operation state of the target machine. The data distribution of each feature dimension in the feature sample group is calculated and the data distribution morphology parameter is recorded. At the same time, the joint distribution relationship between all feature dimensions is calculated and the covariance coefficient matrix between feature dimensions is recorded. The distribution morphology parameter and the covariance coefficient matrix constitute a multidimensional probability model describing the normal response state of the target machine, which is the individualized normal model.

[0017] In S5, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension, specifically including: Each feature component value in the standardized individualized sample feature vector of the current hit event is compared with the historical distribution morphology parameters of the corresponding feature dimension recorded in the individualized normal model. The distance of the feature component value from the center of the historical distribution is calculated, and the distance is converted into a score representing the degree of deviation of a single dimension. The above calculation is performed for all feature dimensions to obtain a set of deviation score vectors of the same dimension as the individualized sample feature vector.

[0018] S5 comprehensively assesses the degree of deviation of each feature dimension and the degree of anomaly of the multi-dimensional joint probability, specifically including: After obtaining the deviation score vectors of each feature dimension, the covariance coefficient matrix between feature dimensions recorded in the individualized normal model is combined to calculate the Mahalanobis distance of the current individualized sample feature vector in the entire multidimensional feature space relative to the overall distribution position of the historical non-abnormal feature sample group. The Mahalanobis distance is compared with a distance threshold obtained from the statistics of historical non-abnormal data. If the distance threshold is exceeded, the target machine response state corresponding to the current hit event is determined to be abnormal in terms of joint probability.

[0019] Further explanation is needed regarding the transient impact vibration signal acquisition process during target drone firing missions. This is achieved through a piezoelectric vibration sensor installed in the middle of the target drone's transmission horizontal shaft. The sensor's sampling frequency is set to 100 kHz. The acquisition trigger signal is output by the target drone's hit detection module. When the hit detection module detects a projectile impacting the target plate, it immediately triggers the vibration sensor to start acquisition, continuously collecting signal data for one second to form a transient impact vibration signal sequence uniquely associated with this hit event. A first time period is defined as a sliding time window of 1 millisecond in length, with a sliding step size of 0.1 milliseconds, used to scan the entire acquired signal segment by segment to detect abrupt changes. A second time period is defined as the total 1-second acquisition duration corresponding to this hit event, used to calculate the global energy characteristics of the signal. After the transient impact vibration signal acquisition for each hit event is completed, a sliding window traversal process is initiated. Starting from the initial moment of signal acquisition, the sliding window is moved sequentially to cover the entire signal sequence. For each sliding window corresponding to the first time period, the local energy within that time period is calculated. The calculation process is as follows: The signal amplitude of all sampling points within the time period is squared, and then all squared values ​​are summed to obtain the local energy corresponding to the window. Simultaneously, the global average energy within the preset second time period is calculated. The calculation process is as follows: the signal amplitude of all sampling points within a 1-second duration is squared, and all squared values ​​are summed to obtain the global total energy. The global total energy is then divided by the total number of sampling points within a 1-second duration to obtain the global average energy. The local energy of each sliding window is divided by the global average energy to obtain the energy ratio corresponding to that window. The target machine's historical no-abnormal-hit event data refers to the transient impact vibration signal data set corresponding to all hit events of the same target machine during the factory commissioning phase and in early target firing missions without mechanical failures, after manual verification confirming that the target plate was impacted normally, the transmission mechanism was not jammed, and the lifting and lowering actions were normal. The dynamic threshold is obtained by statistically analyzing the target machine's historical no-abnormal-hit event data. The specific process is as follows: Extract the energy ratio corresponding to each sliding window in all historical anomaly-free hit events, find the maximum energy ratio in each hit event, calculate the arithmetic mean of all maximum values, and multiply this mean by 1.2 to use as the initial dynamic threshold. After every 100 verified anomaly-free hit events, automatically add new anomaly-free data to the statistical set, recalculate and update the dynamic threshold to ensure that the threshold adapts to the mechanical characteristics changes of the target machine after long-term use. False mutation points are determined using a dual-condition joint judgment method. The first condition is that the energy ratio of the current sliding window exceeds the dynamic threshold, and the second condition is that the rate of change of the signal amplitude within the current sliding window exceeds the preset rate of change threshold. The calculation process of the rate of change is as follows: The maximum signal amplitude within the window is subtracted from the minimum amplitude, and then divided by the window duration of 1 millisecond. Only when both conditions are met simultaneously can a false abrupt change be determined within the preset first time period. The center time of the sliding window is then marked as the false abrupt change point. The predetermined region for the false abrupt change point is defined as a continuous time region extending 0.5 milliseconds before and after the false abrupt change point, centered on the time axis. This region covers the entire range of influence of the false abrupt change point on the signal. After determining the false abrupt change point and defining the predetermined region, an interpolation replacement operation is performed. First, determine the start and end times of the predetermined region for false mutation points. Extract the signal segment within 1 millisecond before the start time as the forward effective impulse waveform segment, and extract the signal segment within 1 millisecond after the end time as the backward effective impulse waveform segment. Read the signal amplitude of the last sampling point of the forward effective impulse waveform segment and the signal amplitude of the first sampling point of the backward effective impulse waveform segment, respectively. Calculate the replacement amplitude of each sampling point within the predetermined region using linear interpolation. The calculation process is as follows: For any sampling point within the predetermined area, its replacement amplitude is equal to the amplitude of the last sampling point of the forward effective impact waveform segment, plus the difference between the amplitude of the first sampling point of the backward effective impact waveform segment and the amplitude of the last sampling point of the forward effective impact waveform segment, multiplied by the time length from the sampling point to the start time of the predetermined area, and then divided by the total time length of the predetermined area (1 millisecond). The calculated replacement amplitude is used to replace the original signal amplitude of the corresponding sampling points within the predetermined area one by one, thus eliminating false abrupt changes. For example, in a hit event, a target drone detects a false abrupt change at 0.2 seconds, corresponding to a predetermined area of ​​0.199. From 5 seconds to 0.2005 seconds, the amplitude of the last sampling point of the forward effective impact waveform segment is 0.05 volts, and the amplitude of the first sampling point of the backward effective impact waveform segment is 0.06 volts. The sampling point at 0.2 seconds in the predetermined area has a time length of 0.0005 seconds from the start time. Its replacement amplitude is 0.05 volts plus 0.01 volts multiplied by 0.0005 seconds divided by 0.001 seconds, which is 0.055 volts. After replacement, the signal amplitude in this area achieves a smooth transition without abrupt spikes. The transient impact vibration signal after eliminating false abrupt points is used for the identification of the peak moment of subsequent impact energy and the truncation operation of impact waveform segments.

[0020] After eliminating false abrupt change points in the transient impact vibration signal input, all sampling points of the signal are read. The signal sampling frequency is 100 kHz, and the corresponding sampling interval is 10 microseconds. Starting from the beginning of signal acquisition, the square of the signal amplitude at each sampling point is calculated to obtain the instantaneous power value corresponding to each sampling point. Then, discrete-time integration is performed on the instantaneous power value to generate the impact energy accumulation curve. The specific process of discrete integration is as follows: Starting from the first sampling point, the instantaneous power value of the current sampling point is multiplied by the sampling interval, and then added to the cumulative energy value corresponding to the previous sampling point to obtain the cumulative impact energy value of the current sampling point. After calculating the cumulative energy values ​​of all sampling points in sequence, the time of each sampling point is used as the abscissa and the corresponding cumulative impact energy value is used as the ordinate to plot a continuous impact energy accumulation curve. The instantaneous slope of each sampling point on the impact energy accumulation curve is calculated using the sliding window method. The length of the sliding window is set to 11 sampling points, that is, including the current sampling point and 5 sampling points before and after it. The calculation process of the instantaneous slope is as follows: The instantaneous slope value of the current sampling point is obtained by subtracting the cumulative energy value of the first sampling point from the cumulative energy value of the last sampling point within the window, and then dividing by the total time length of the window, 100 microseconds. Predefined criteria are used to determine whether the slope is continuously and rapidly increasing or transitioning to a gradual change. Continuous rapid increase is defined as 15 or more consecutive sampling points having instantaneous slope values ​​greater than a rapid slope threshold. The rapid slope threshold is obtained by statistically analyzing the slope of the impact energy accumulation curve of historical target machine events without abnormal hits. Specifically: Extract the maximum instantaneous slope value during the rising phase of the impact response from all historical non-anomaly hit events. Calculate the arithmetic mean of all maximum values. Set 80% of this mean as the rapid slope threshold. A transition to a gradual change is defined as when the instantaneous slope values ​​of 10 or more consecutive sampling points are all less than the gradual slope threshold. The gradual slope threshold is set to 20% of the rapid slope threshold. The inflection point is defined as the moment when the slope on the impact energy accumulation curve first transitions from meeting the condition of continuous rapid increase to meeting the condition of gradual change. Starting from the beginning of the impact energy accumulation curve, scan the instantaneous slope value point by point, recording the starting position where the slope of the first 15 consecutive sampling points is greater than the rapid slope threshold. Continue scanning forward. When the slope of the first 10 consecutive sampling points is detected to be less than the gradual slope threshold, the time corresponding to the first sampling point in those 10 consecutive sampling points is determined as... The inflection point is the peak value of the impact energy in the main waveform of the impact response. For example, the rapid slope threshold obtained from the historical anomaly-free data of a target machine is 60 joules per second, and the gradual slope threshold is 12 joules per second. In the impact energy accumulation curve of a certain hit event, it is detected that the slope of 15 consecutive sampling points from 0.1200 seconds to 0.1214 seconds is greater than 60 joules per second, which meets the condition of continuous rapid increase. When scanning to 0.1230 seconds, it is found that the slope of 10 consecutive sampling points from 0.1230 seconds to 0.1239 seconds is less than 12 joules per second, which meets the condition of transitioning to gradual change. Therefore, 0.1230 seconds is determined as the peak value of the impact energy of this hit event. This moment will be used as the alignment reference for subsequent impact waveform segments and current waveform segments to ensure the time synchronization of the two types of signals.

[0021] The fixed time length is determined by statistically analyzing the duration of the main waveform of the impact response of all historical no-abnormal-hit events of the same target drone. The specific process is as follows: The duration of the main waveform of the impact response is extracted for each no-abnormal-hit event. The maximum value of all durations is calculated, and this maximum value is multiplied by 1.2 to obtain a fixed duration, which is set to 20 milliseconds in this invention. After obtaining the peak moment of the impact energy, the window start time is obtained by extending 20 milliseconds forward from the time axis and the window end time is obtained by extending 20 milliseconds backward from the time axis. The transient impact vibration signal between the window start time and the window end time is extracted and defined as an impact waveform segment. The peak amplitude of the impact response is extracted within the impact waveform segment. The specific process is as follows: Iterate through all sampling points within the impact waveform segment, calculate the absolute value of the signal amplitude at each sampling point, compare the magnitudes of all absolute values, and determine the maximum value as the peak amplitude of the impact response. For example, if the maximum absolute value of the signal amplitude at a sampling point within a certain impact waveform segment is 2.5 volts, then the peak amplitude of the impact response for this event is 2.5 volts. Next, perform the extraction of the impact energy integral. The specific process is as follows: Traverse all sampling points within the impact waveform segment, calculate the square of the signal amplitude at each sampling point, multiply each square by the signal sampling interval of 10 microseconds, and then sum all the products. The sum is the integral of the impact energy. For example, if an impact waveform segment contains 4000 sampling points, and the sum of the squares of all sampling points multiplied by 10 microseconds is 0.012 joules, then the integral of the impact energy of this hit event is 0.012 joules. The noise threshold is determined by statistically analyzing the background vibration signal under the condition of no hit by the target. The specific process is as follows: Background vibration signals were collected for 10 seconds when the target drone was stationary and without impact. The standard deviation of the amplitude at all sampling points of this signal was calculated, and three times the standard deviation was set as the noise threshold. The impact response duration was then extracted. The specific process is as follows: Starting from the beginning of the window of the impact waveform segment, the signal amplitude is scanned point by point. The time corresponding to the first sampling point where the signal amplitude exceeds the noise threshold is recorded, and this time is marked as the start time of the impact response. Then, starting from the end of the window of the impact waveform segment, the signal amplitude is scanned point by point backward, and the time corresponding to the last sampling point where the signal amplitude exceeds the noise threshold is recorded, and this time is marked as the end time of the impact response. The duration of the impact response is the time length obtained by subtracting the start time of the impact response from the end time of the impact response. For example, if the noise threshold of a target machine is 0.02 volts, the time when the noise threshold is first exceeded in the impact waveform segment is 0.1130 seconds, and the time when it falls back below the noise threshold is 0.1320 seconds, then the duration of the impact response of this hit event is 0.019 seconds. The extracted peak amplitude of the impact response, the integral of the impact energy, and the duration of the impact response are arranged in the above order to form an impact feature vector. This impact feature vector will be concatenated with the subsequently extracted electromechanical response feature vector to form an individualized sample feature vector used to construct an individualized normal model.

[0022] After acquiring the current waveform segment aligned with the time of the impact waveform segment, the threshold for determining the zero current value is determined. The threshold for determining the zero current value is calculated by acquiring the background current signal for 10 seconds in the state where the target drone's tilting motor is not driven. The specific process is as follows: The standard deviation of the amplitude of all sampling points of the background current in this segment is calculated. Three times the standard deviation is set as the current zero-value judgment threshold. The current sampling frequency is set to 10 kHz, corresponding to a sampling interval of 100 microseconds. The system scans the current sampling values ​​point by point starting from the beginning of the current waveform segment, identifying the starting point of each current rise from zero. The rule for determining the starting point is that the current value of three consecutive sampling points is greater than the current zero-value judgment threshold, and the current value of the subsequent sampling point is greater than the current value of the previous sampling point. The time corresponding to the first sampling point among these three consecutive sampling points that exceeds the current zero-value judgment threshold is marked as the current rise starting point, and the process continues. The subsequent scan identifies the end point where the current falls back to zero. The rule for determining the end point is that the absolute value of the current at three consecutive sampling points is less than the current zero-value threshold, and the current value at the next sampling point is less than the current value at the previous sampling point. The time corresponding to the last sampling point among the three consecutive sampling points that exceeds the current zero-value threshold is marked as the end point of the current fallback. The time length between two adjacent current rise start points is defined as a complete start-up / downward drive cycle. All start-up / downward drive cycles contained in the current waveform segment are divided sequentially. For each segmented start-up / downward drive cycle, the current rise phase within that cycle is determined. The specific process is as follows: Iterate through all current sampling points within the current reverse drive cycle, find the time corresponding to the sampling point with the largest current value, and determine all sampling points between the current rise start point and the time of the maximum value as the data points of the current rise phase. Use the least squares method to perform linear fitting on the data points of the current rise phase. The fitting process is as follows: First, count the total number of sampling points during the current rise phase, denoted as n. Use the time corresponding to each sampling point as the x-axis and the corresponding current value as the y-axis. Calculate the sum of all x-coordinates, the sum of all y-coordinates, the sum of the squares of all x-coordinate values, and the sum of the products of all x-coordinate values ​​and their corresponding y-coordinate values. Then, calculate the slope of the fitted line. The slope is calculated by multiplying n by the sum of the products of the x-coordinates and y-coordinates, subtracting the sum of the products of the x-coordinates and y-coordinates to get the numerator; multiplying n by the sum of the squares of the x-coordinates, subtracting the square of the sum of the squares of the x-coordinates to get the denominator; and dividing the numerator by the denominator. The result is the current rise slope of that inverted drive cycle. For example, in a certain inverted drive cycle, the current rise phase contains 50 sampling points, the sum of all x-coordinates is 25 milliseconds, the sum of all y-coordinates is 20 amperes, the sum of the squares of all x-coordinates is 15 square milliseconds, and the sum of the products of all x-coordinates... The total is 120 amperes per millisecond. Therefore, the current rise slope is calculated as the result of 50 multiplied by 120 minus 25 multiplied by 20 divided by the square of 50 multiplied by 15 minus 25. The final calculated current rise slope is 4 amperes per millisecond. The time length obtained by subtracting the current rise start point from the next current rise start point corresponding to this inverted drive cycle is taken as the inverted drive cycle. The current integral of this inverted drive cycle is calculated by iterating through all current sampling points within the cycle, calculating the absolute value of the current value at each sampling point, multiplying each absolute value by the current sampling interval of 100 microseconds, and then summing all the products. The sum is the current integral of this inverted drive cycle. After extracting the current rise slope, inverted drive cycle, and current integral for all inverted drive cycles, the time overlap between each inverted drive cycle and the impact waveform segment is calculated. The overlap calculation process is as follows: Calculate the intersection length between the time interval of the initiation-reverse drive cycle and the time interval of the impact waveform segment. Divide the intersection length by the total time length of the impact waveform segment to obtain the time overlap value of the initiation-reverse drive cycle. Compare the overlap values ​​of all initiation-reverse drive cycles and find the initiation-reverse drive cycle with the largest value. Determine this cycle as the initiation-reverse drive cycle with the highest time overlap with the impact waveform segment. Arrange the current rise slope, initiation-reverse cycle, and current integral value extracted from the initiation-reverse drive cycle with the highest overlap in the above order to form an electromechanical response feature vector. This electromechanical response feature vector will be concatenated with the impact feature vector to form an individualized sample feature vector used to construct an individualized normal model of the target machine.

[0023] Before constructing a unique, individualized normal model for the target drone, all hit event data from all historical, mechanically abnormal states of the same target drone are retrieved. From this data, the impact feature vector and electromechanical response feature vector corresponding to each hit event are extracted. The impact feature vector includes three feature components: peak impact response amplitude, impact energy integral, and impact response duration. The electromechanical response feature vector includes three feature components: current rise slope, start-up / reset period, and current integral. The historical average and historical standard deviation are calculated for each of the three feature components of the impact feature vector. The calculation process for the historical average of a single feature component is as follows: The historical standard deviation for a single feature component is calculated by summing the values ​​of all historical anomaly-free data and then dividing the sum by the total number of historical anomaly-free hit events. The difference is obtained by subtracting the corresponding historical average value from the historical value of each historical anomaly-free data point of the feature component. All differences are squared and summed. The sum is divided by the total number of historical anomaly-free hit events to obtain the variance. The square root of the variance is then taken to obtain the historical standard deviation. The same calculation method is used to calculate the corresponding historical average and historical standard deviation for each of the three feature components of the electromechanical response feature vector. The historical average and historical standard deviation of the six feature components are stored in the target drone's dedicated feature parameter library for subsequent feature standardization processing of single hit events. For any hit event during the target drone's firing mission, the impact feature vector and electromechanical response feature vector corresponding to the hit event are first extracted. Then, the historical average and historical standard deviation corresponding to each feature component are retrieved from the feature parameter library. Standardization processing is then performed on each feature component sequentially. The standardization process is as follows: The standardized value of a feature component is obtained by subtracting its historical average value from the current value of that component and then dividing the difference by the historical standard deviation of that component. For example, if the historical average value of the peak amplitude of the impact response of a target drone is 2.5 volts and the historical standard deviation is 0.3 volts, and the peak amplitude of the impact response of the current hit event is 2.8 volts, then the standardized value of this component is 2.8 volts minus 2.5 volts, divided by 0.3 volts, resulting in 1. After the system completes the standardization of the three feature components of the impact feature vector and the three feature components of the electromechanical response feature vector, it connects them in a fixed order: the standardized value of the peak amplitude of the impact response, the standardized value of the impact energy integral, the standardized value of the impact response duration, the standardized value of the current rise slope, the standardized value of the start-up / reversal period, and the standardized value of the current integral. This six standardized feature values ​​are combined into a six-dimensional feature sequence, which is the individualized sample feature vector corresponding to the current hit event. The individualized sample feature vector will serve as the basic data for constructing the individualized normal model of the target drone and will also serve as the input data for judging abnormal states in subsequent target firing missions.

[0024] The standardized individualized sample feature vectors corresponding to all historical hit events of the same target machine without mechanical abnormalities are aggregated and combined to form an individualized sample feature vector set. This set directly serves as the feature sample group under normal operating conditions of the target machine. The feature sample group contains six feature dimensions: standardized impact response peak amplitude, standardized impact energy integral, standardized impact response duration, standardized current rise slope, standardized start-up / reset period, and standardized current integral. The data distribution corresponding to each feature dimension in the feature sample group is calculated independently. The calculation process for the data distribution of a single feature dimension is as follows: Extract all standardized feature values ​​for this dimension, first calculating the mean of this dimension, as follows: First, sum all the values ​​for this dimension and divide by the total number of samples in the feature sample group. Then, calculate the variance of this dimension by subtracting the corresponding mean from each value in this dimension, summing all the squared differences, and then dividing by the total number of samples. Next, calculate the skewness of this dimension by summing the cubes of the differences between each value in this dimension and the mean, dividing by the total number of samples, and then dividing by the cube root of the variance. Finally, calculate the kurtosis of this dimension. Subtract the mean from each value in this dimension to obtain the fourth power of the difference, sum these differences, divide by the total number of samples, and then divide by the square of the variance. The data distribution shape parameter is defined as the statistical characteristic constant of the data distribution of a single feature dimension, specifically including four types of parameters for each feature dimension: mean, variance, skewness, and kurtosis. Record and store the four types of parameters for each of the six feature dimensions one by one, and calculate the joint distribution relationship between all pairs of feature dimensions. The joint distribution relationship refers to the degree of statistical correlation when the standardized values ​​of two feature dimensions change simultaneously. It is used to characterize the coordinated change law between different response characteristics under normal target machine conditions. The core of calculating the joint distribution relationship between pairwise feature dimensions is the covariance. The calculation process of covariance is as follows: Extract all values ​​for each of the two feature dimensions. Subtract the mean from each value in the first dimension and the mean from each value in the second dimension. Multiply the two sets of differences and sum them. Then divide by the total number of samples. Calculate the covariance between any two dimensions in each of the six feature dimensions, resulting in fifteen sets of covariance values. The covariance coefficient matrix between the feature dimensions is a 6x6 square matrix, with each row and column corresponding to one of the six feature dimensions. The elements on the diagonal are the variance of each feature dimension itself, and the elements off-diagonal are the covariance between the corresponding two feature dimensions. Fill the calculated variances and covariances into the corresponding positions in the square matrix according to the dimensional order to complete the construction and recording of the covariance coefficient matrix. The construction process of the multidimensional probability model is as follows: The mean, variance, skewness, and kurtosis of six feature dimensions are used as the basis for single-dimensional probabilistic description, and the six-row, six-column covariance coefficient matrix is ​​used as the basis for multi-dimensional joint probabilistic description. The two types of data are integrated into a unified probabilistic description system. This system can completely characterize the single-dimensional distribution law and multi-dimensional collaborative change law of each response feature of the target machine under the condition of no mechanical anomaly. This integrated probabilistic description system is the individualized normal model unique to the target machine. For example, the total number of samples in the feature sample group of a certain target machine is 500. The mean of the standardized dimension of the peak amplitude of the impact response is zero, the variance is one, the skewness is 0.1, and the kurtosis is 3.0. The covariance between this dimension and the standardized dimension of the current rise slope is 0.2. The elements in the first row, second column and the second row, first column of the covariance coefficient matrix are all 0.2, and the elements on the diagonal are the variances of each dimension. The data distribution morphology parameters and the covariance coefficient matrix together constitute the individualized normal model of the target machine, which is used for subsequent anomaly determination of the hit event.

[0025] Obtain the standardized six-dimensional individualized sample feature vector of the current hit event. The six feature components of this vector correspond in order to the standardized dimensions of the peak amplitude of the impact response, the integrated impact energy, the duration of the impact response, the current rise slope, the start-up / fallback period, and the integrated current. Retrieve the data distribution morphology parameters corresponding one-to-one with the six feature dimensions from the individualized normal model already constructed on the target machine. The data distribution morphology parameters include the historical mean, historical variance, historical skewness, and historical kurtosis for each dimension. Perform calculation operations on each feature component separately in dimensional order. First, calculate the distance of the current feature component value from the historical distribution center of the corresponding dimension. The calculation method is as follows: Subtract the historical mean of that dimension from the current feature component value, and take the absolute value of the difference. This absolute value is the distance of a single dimension from the historical distribution center. Then, convert this deviation distance into a score representing the degree of deviation of the single dimension. The conversion method is as follows: The deviation score is obtained by dividing the deviation distance by the square root of the historical variance of that dimension. The system repeats the above deviation distance calculation and deviation score conversion operation for all six feature dimensions. For example, if the standardized component value of the peak amplitude of the impact response of the current hit event is 1.5, the historical mean of this dimension is 0, the historical variance is 1, and the square root of the historical variance is 1, then the deviation distance is 1.5 and the deviation score is 1.5. As another example, if the standardized component value of the start-up period is -0.6, the historical mean of this dimension is 0, the historical variance is 1, the deviation distance is 0.6, and the deviation score is 0.6. The system arranges the deviation scores calculated from the six dimensions in a fixed order: peak amplitude of impact response, impact energy integral, impact response duration, current rise slope, start-up period, and current integral. Finally, a set of six-dimensional deviation score vectors with the same number of dimensions as the individualized sample feature vectors is generated. This deviation score vector will serve as the basis for calculating the Mahalanobis distance in combination with the covariance coefficient matrix in the individualized normal model, and will be used to complete the multi-dimensional joint probability anomaly determination of the target machine response state.

[0026] After obtaining the six-dimensional deviation score vector corresponding to the current hit event, the pre-recorded six-row, six-column covariance coefficient matrix among the feature dimensions is retrieved from the target machine's individualized normal model. The inverse operation of this covariance coefficient matrix is ​​then performed to obtain the inverse covariance coefficient matrix. The inverse operation is completed using the adjoint matrix method. The specific process is as follows: Calculate the determinant of the covariance coefficient matrix, then calculate the algebraic cofactor of each element in the matrix to form the adjoint matrix. Divide the adjoint matrix by the determinant to obtain the inverse covariance coefficient matrix. Finally, calculate the Mahalanobis distance. The calculation process is divided into: The six-dimensional deviation fractional vector is converted into a six-row, one-column column vector. The column vector is then transposed to obtain a six-row, one-column row vector. This transposed row vector is multiplied by the inverse matrix of the covariance coefficients, and the product is multiplied by the original six-dimensional column vector to obtain a single-valued value. The square root of this single-valued value is then taken to obtain the Mahalanobis distance of the current individualized sample feature vector relative to the overall distribution of the historical non-abnormal feature sample group in the six-dimensional feature space. The distance threshold is defined as the Mahalanobis distance values ​​of the individualized sample feature vectors corresponding to all historical non-abnormal mechanical state hit events of the target machine. All Mahalanobis distance values ​​are sorted from smallest to largest, and the value corresponding to the 99th percentile is selected. This value is the critical distance value for the joint distribution of multi-dimensional features under normal target machine conditions. The target machine response state refers to the comprehensive electromechanical response state constituted by the vibration response of the target plate transmission axis and the drive response of the tilting motor after the target machine is hit by a projectile. The joint probability refers to the probability that the response values ​​of the six feature dimensions are... The probability that the current individualized sample feature vector conforms to the single-dimensional distribution law and the collaborative change law between dimensions in the individualized normal model is used to characterize the normal probability of the state under the collaborative effect of multiple features. The calculated Mahalanobis distance is compared with the distance threshold. If the Mahalanobis distance is greater than the distance threshold, it means that the current individualized sample feature vector is far away from the distribution center of the historical abnormal sample group in the multi-dimensional feature space. The six feature dimensions cannot simultaneously satisfy the collaborative relationship of normal distribution. The multi-feature joint probability of the current hit event is lower than the critical probability of the normal state. That is, the target machine response state corresponding to the current hit event is determined to be abnormal in terms of joint probability. For example, if the distance threshold of a target machine is set to 3.0, and the Mahalanobis distance calculated for the current hit event is 3.8, 3.8 is greater than 3.0. It is determined that the comprehensive electromechanical response state of the target machine corresponding to this hit is abnormal and the subsequent abnormal alarm process needs to be triggered. If the Mahalanobis distance is less than or equal to the distance threshold, it is determined that the current target machine response state conforms to the normal joint probability distribution and the state is not abnormal.

[0027] The second objective of this invention is to provide a system for implementing a big data-based intelligent monitoring method for target drone shooting, including any of the above-mentioned methods, comprising: The signal synchronization acquisition unit is used to synchronously acquire the transient impact vibration signal transmitted through the transmission horizontal shaft after the target plate is impacted, as well as the current waveform of the tilting motor during the driving cycle of the hit event, with the time of the hit event as the time reference, during the operation of the target machine, to form an impact vibration sequence and current waveform sequence that are uniquely associated with each hit event. The signal processing and segment extraction unit is used to apply adaptive filtering to the transient impact vibration signal to eliminate false abrupt change points, and to extract the impact waveform segment containing the main waveform of the impact response based on the peak time of the impact energy in the transient impact vibration signal. At the same time, it extracts the current waveform segment that is time-aligned with the impact waveform segment from the current waveform of the starting and stopping motor. The feature extraction and standardization unit is used to extract the peak amplitude of the impact response, the integral amount of the impact energy, and the duration of the impact response from the impact waveform segment to form an impact feature vector, and to extract the current rise slope, the start-up and reversal period, and the integral amount of the current from the current waveform segment to form an electromechanical response feature vector. The impact feature vector and the electromechanical response feature vector are then standardized and concatenated to form an individualized sample feature vector. The model building and anomaly judgment unit is used to construct a unique individualized normal model for the target machine by using the individualized sample feature vectors corresponding to all historical hit events of the same target machine under no mechanical abnormality. In subsequent target shooting tasks, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension and the degree of multi-dimensional joint probability anomaly.

[0028] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A target drone shooting intelligent monitoring method based on big data, characterized in that: Includes the following steps: S1. During the operation of the target machine, the transient impact vibration signal transmitted through the transmission horizontal shaft after the target plate is impacted is synchronously collected, and the current waveform of the tilting motor during the driving cycle of the hit event is collected, taking the moment of the hit event as the time reference, to form an impact vibration sequence and current waveform sequence uniquely associated with each hit event. S2. Apply adaptive filtering to the transient impact vibration signal to eliminate false abrupt change points, and use the peak time of the impact energy in the transient impact vibration signal as the alignment reference to extract the impact waveform segment containing the main waveform of the impact response. At the same time, extract the current waveform segment from the current waveform of the starting and stopping motor that is time-aligned with the impact waveform segment. S3. Extract the peak amplitude of the impact response, the integral of the impact energy, and the duration of the impact response from the impact waveform segment to form an impact feature vector. Extract the current rise slope, the start-up / reversal period, and the integral of the current from the current waveform segment to form an electromechanical response feature vector. Concatenate the impact feature vector and the electromechanical response feature vector to form an individualized sample feature vector, including: Before constructing a unique individualized normal model for the target drone, the historical average and standard deviation of each characteristic component in the impact characteristic vector and the historical average and standard deviation of each characteristic component in the electromechanical response characteristic vector are calculated using all historical hit event data of the target drone under no mechanical abnormality conditions. For any hit event, after obtaining the impact feature vector and the electromechanical response feature vector, the value of each feature component is subtracted from the historical average value corresponding to the feature component, and then divided by the historical standard deviation corresponding to the feature component to complete the standardization process. The standardized impact feature vector and the electromechanical response feature vector are then connected in sequence to form the individualized sample feature vector. S4. Using big data technology, the individualized sample feature vectors corresponding to all historical hit events of the same target drone under normal mechanical abnormality conditions are used as training data to construct a unique individualized normal model for the target drone, including: The individualized sample feature vectors corresponding to all historical events without mechanical abnormality are standardized and then combined to form an individualized sample feature vector set, which is regarded as the feature sample group under the normal operation state of the target machine. The data distribution of each feature dimension in the feature sample group is calculated and the data distribution morphology parameter is recorded. At the same time, the joint distribution relationship between all feature dimensions is calculated and the covariance coefficient matrix between feature dimensions is recorded. The distribution morphology parameter and the covariance coefficient matrix constitute a multidimensional probability model describing the normal response state of the target machine, which is the individualized normal model. S5. In subsequent target practice missions, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension and the degree of abnormality of the multi-dimensional joint probability.

2. The intelligent monitoring method for target drone firing based on big data according to claim 1, characterized in that: The adaptive filtering applied to the transient impact vibration signal in S2 specifically includes: After the transient impact vibration signal corresponding to each hit event is collected, the ratio of the local energy of the transient impact vibration signal in a preset first time period to the global average energy in a preset second time period is calculated. When the ratio exceeds the dynamic threshold obtained from the statistical data of the target machine's historical no abnormal hit events, a false mutation point is determined to exist in the preset first time period. The signal amplitude of the transient impact vibration signal in the predetermined area of ​​the false mutation point is replaced by the signal amplitude of the effective impact waveform segments before and after it to eliminate the false mutation point.

3. The intelligent monitoring method for target drone firing based on big data according to claim 2, characterized in that: The alignment reference in S2 is based on the peak moment of the impact energy in the transient impact vibration signal, specifically including: On the transient impact vibration signal after the elimination of false abrupt change points, the square of the signal amplitude is calculated and integrated over time to obtain the impact energy accumulation curve. The inflection point on the impact energy accumulation curve where the slope changes from a continuous rapid increase to a gradual change is identified, and the time corresponding to this inflection point is determined as the peak time of the impact energy of the main waveform of the impact response.

4. The intelligent monitoring method for target drone firing based on big data according to claim 3, characterized in that: In step S3, the peak amplitude of the impact response, the integral of the impact energy, and the duration of the impact response are extracted from the impact waveform segment to form an impact feature vector, specifically including: Centered on the peak moment of the impact energy, a window of fixed time length is extended to both sides of the time axis. The signal within the window is defined as an impact waveform segment. Within the impact waveform segment, the maximum absolute value of the signal is taken as the peak amplitude of the impact response. The result of integrating the square of the signal amplitude over the time range of the impact waveform segment is taken as the impact energy integral. The time length from when the signal amplitude first exceeds the noise threshold to when it finally falls back below the noise threshold within the impact waveform segment is taken as the impact response duration. The peak amplitude of the impact response, the impact energy integral, and the impact response duration are arranged in order to form an impact feature vector.

5. The intelligent monitoring method for target drone firing based on big data according to claim 1, characterized in that: In step S3, the current rise slope, start-up / reversal period, and current integral are extracted from the current waveform segment to form an electromechanical response feature vector, specifically including: Within the current waveform segment, the starting point of each current rise from zero and the ending point of the current fall back to zero are identified. The time length between one starting point and the next starting point is defined as a complete start-down drive cycle. Within a single start-down drive cycle, the data points of the current rise phase are fitted with a straight line. The slope of the resulting straight line is taken as the current rise slope of a single start-down drive cycle. The time length of a single start-down drive cycle is taken as the start-down cycle. The result of integrating the absolute value of the current within the time range of a single start-down drive cycle is taken as the current integral. From the multiple start-down drive cycles contained in the current waveform segment, the start-down drive cycle with the highest temporal overlap with the impact waveform segment is selected. The current rise slope, start-down cycle, and current integral extracted within the highest start-down drive cycle are arranged in order to form an electromechanical response feature vector.

6. The intelligent monitoring method for target drone firing based on big data according to claim 1, characterized in that: In step S5, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension, specifically including: Each feature component value in the standardized individualized sample feature vector of the current hit event is compared with the historical distribution morphology parameters of the corresponding feature dimension recorded in the individualized normal model. The distance of the feature component value from the historical distribution center is calculated, and the distance is converted into a score representing the degree of deviation in a single dimension. The above calculation is performed for all feature dimensions to obtain a set of deviation score vectors of the same dimension as the individualized sample feature vector.

7. The intelligent monitoring method for target drone firing based on big data according to claim 1, characterized in that: The S5 process comprehensively judges the degree of deviation of each feature dimension and the degree of anomaly of the multi-dimensional joint probability, specifically including: After obtaining the deviation score vectors of each feature dimension, the covariance coefficient matrix between feature dimensions recorded in the individualized normal model is combined to calculate the Mahalanobis distance of the current individualized sample feature vector in the entire multidimensional feature space relative to the overall distribution position of the historical non-abnormal feature sample group. The Mahalanobis distance is compared with a distance threshold obtained from the statistics of historical non-abnormal data. If the distance threshold is exceeded, the target machine response state corresponding to the current hit event is determined to be abnormal in terms of joint probability.

8. A system for implementing a big data-based intelligent monitoring method for target drone shooting, as described in any one of claims 1-7, characterized in that, include: The signal synchronization acquisition unit is used to synchronously acquire the transient impact vibration signal transmitted through the transmission horizontal shaft after the target plate is impacted, as well as the current waveform of the tilting motor during the driving cycle of the hit event, with the time of the hit event as the time reference, during the operation of the target machine, to form an impact vibration sequence and current waveform sequence that are uniquely associated with each hit event. The signal processing and segment extraction unit is used to apply adaptive filtering to the transient impact vibration signal to eliminate false abrupt change points, and to extract an impact waveform segment containing the main waveform of the impact response based on the peak time of the impact energy in the transient impact vibration signal. At the same time, it extracts a current waveform segment from the current waveform of the starting and stopping motor that is time-aligned with the impact waveform segment. The feature extraction and standardization unit is used to extract the peak amplitude of the impact response, the integral amount of the impact energy, and the duration of the impact response from the impact waveform segment to form an impact feature vector, and to extract the current rise slope, the start-up and reversal period, and the integral amount of the current from the current waveform segment to form an electromechanical response feature vector. The impact feature vector and the electromechanical response feature vector are then standardized and concatenated to form an individualized sample feature vector. The model building and anomaly judgment unit is used to construct a unique individualized normal model for the target machine using the individualized sample feature vectors corresponding to all historical hit events of the same target machine under no mechanical abnormality conditions. In subsequent target shooting tasks, the individualized sample feature vector corresponding to the current hit event is input into the individualized normal model to comprehensively judge the degree of deviation of each feature dimension and the degree of multi-dimensional joint probability anomaly.

Images