Multi-channel pulse extraction and matching method and device for positioning redundant objects in sealed electronic equipment

Abstract

This invention relates to a multi-channel pulse extraction and matching method and device for locating foreign objects in sealed electronic equipment, belonging to the field of foreign object monitoring technology for sealed electronic equipment. To address the problems of inaccurate pulse extraction for continuous multi-pulse processing in existing multi-channel pulse extraction and matching methods, and matching errors due to missing values ​​or abnormal symbols in some channels, this invention first employs a two-stage variable multi-threshold pulse extraction algorithm. Based on the 3σ distribution criterion of a normal distribution, peak energy threshold, endpoint energy threshold, and endpoint zero-crossing rate threshold are set in the first stage. In the second stage, the endpoint energy threshold and endpoint zero-crossing rate threshold are reset and detected. The true start and end times of the current pulse are determined based on the signal duration ratio of the two stages. Then, using the pulse in the foreign object signal of the channel furthest from the foreign object as a reference, a threshold-based pulse matching algorithm is used for multi-channel matching.

Description

Technical Field

[0001] The invention belongs to the technical field of monitoring excess matter in sealed electronic equipment, and in particular relates to a pulse extraction and matching method for excess matter signals in sealed electronic equipment, and corresponding equipment. Background Art

[0002] Sealed electronic devices (SSE) are complex systems composed of mechanical structures, integrated circuits, and actuators. They have functions such as sealing, waterproofing, and electromagnetic shielding, and are widely used in the field of aerospace engineering. Due to the limitations of the production process, particles such as solder particles, metal gaskets, and wire segments may be encapsulated in the production process of sealed electronic devices. These substances that enter from the outside or are generated inside and do not conform to the specified stable physical state of sealed electronic devices are called excess matter. Sealed electronic devices usually operate in a super-weightless environment, and the excess matter inside will be activated and in a state of random motion. They will collide with internal components to damage them, attach to the circuit surface to cause short circuits and open circuits, etc., posing a major threat to the reliable operation of sealed electronic devices. The particle impact noise detection (PIND) method is a widely verified and effective excess matter detection method. It uses the principle of vibration sound generation to detect whether sound is generated inside the sealed electronic device to determine whether there is excess matter. At present, the PIND detection equipment invented based on this method is widely used worldwide.

[0003] For sealed electronic equipment, once the excess material is detected, it needs to be opened to find and clean the excess material. Sealed electronic equipment is heavy and large in size, and its internal structure is complex. Finding and cleaning the excess material requires a lot of resources. It is very important to obtain the location of the excess material, which can provide a reference for the excess material cleaning work. The excess material in a random motion state will collide or slide to produce sound, which is captured by the acoustic emission sensor of the PIND detection equipment and converted into an electrical signal, called the excess material signal. In terms of the generation mechanism, the excess material signal belongs to the acoustic emission signal in a broad sense. Therefore, some scholars use the time difference positioning method and the regional positioning method to carry out the research on the positioning of excess materials. However, due to the complex internal structure and uneven composition of sealed electronic equipment, the positioning results obtained by these methods are inaccurate and unstable, and have not been applied in real application scenarios. The research on the positioning of excess materials was once stagnant. In recent years, with the development and application of machine learning, the signal recognition method based on machine learning has become mature, and has achieved good application results in fields such as fault diagnosis, providing a reference for the research on the positioning of excess materials. There are differences in acoustic properties between different types of signals, and these differences can be quantified as differences in distribution characteristics between data with different labels in the data set. Classifiers can be used to capture these differences in high-order feature space and set appropriate hyperplanes to complete data classification, thereby achieving signal recognition. Similarly, there are differences in acoustic properties between the redundant signals generated at different locations inside sealed electronic devices. By constructing a data set from the signals and training appropriate classifiers, it is possible to identify the redundant signals generated at different locations. This is redundant object localization.

[0004] In previous studies, Sun Zhigang et al. first proposed to transform the problem of redundant object positioning into a classification problem of machine learning. For the redundant object signals generated at different locations, 11 signal features were extracted from the time domain and frequency domain to construct a data set, and a random forest with optimized parameters was trained to achieve a classification accuracy of 84.08%. Subsequently, Sun Zhigang et al. conducted in-depth research from multiple aspects to improve the classification accuracy, including selecting the latest classifiers with good performance, using optimization algorithms to find the optimal parameter combination of the classifier, processing incomplete data and optimizing features of the data set, and extracting new signal features from multiple domains. Although the classification accuracy was once improved to about 95%, this was only achieved on a fixed research object. When these methods or classifiers were applied to test new sealed electronic devices, even if it had the same properties as the sealed electronic devices used for research, the classification accuracy achieved was not satisfactory. In fact, the signal is the source of this series of studies. The better the quality of the signal, the higher the quality of the constructed data set, and the better the classification performance of the trained classifier. However, in previous studies, only the quality of the data set and the classification performance of the classifier were improved, while the signal was ignored.

[0005] The redundant signal collected and stored by the PIND detection equipment is a digital signal that has been processed by amplification, filtering, A / D sampling, etc., and no additional signal processing is required. In this way, the processing process from signal to data set only has two steps: pulse extraction and feature extraction. Every time the redundant object collides or slides on the inner wall or internal component surface of the sealed electronic device, a peak-shaped pulse will be formed in the redundant signal. When the redundant object is suspended in the air without contacting these materials, a background noise similar to a horizontal plane will be formed in the redundant signal. Therefore, the main components of the redundant signal are pulses and background noise. As mentioned above, the difference in acoustic characteristics of the redundant signals generated at different positions can be quantified as the difference in distribution characteristics between the data. The background noise in the redundant signal is naturally generated and has nothing to do with the redundant object. The pulse is generated by the collision or sliding of the redundant object, is related to the redundant object, and can feedback the difference in acoustic characteristics of the redundant signal generated by the redundant object at different positions. The redundant signal is a digital signal, which is composed of a large number of sampling points arranged at fixed time intervals. The feature extraction process is to calculate the characteristic values ​​of the signal features in multiple domains at the sampling points, construct feature vectors, and obtain data. Therefore, only the data obtained by feature extraction at the sampling points contained in the pulse is meaningful. In previous studies, the work on calculating the characteristic values ​​of which signal features has been completed, that is, the feature extraction step has been determined. Therefore, the process of pulse extraction still needs further research, which becomes the core of the present invention to improve the classification accuracy starting from the source signal.

[0006] At present, there are few studies on pulse extraction algorithms. The three-threshold pulse extraction algorithm is the most commonly used in the field of speech recognition. It calculates the average energy of all sampling points in the signal to set the start detection threshold, peak detection threshold and end detection threshold, that is, three thresholds. Then, the peak detection threshold is used to find the highest point of the pulse, and then the start and end points of the pulse are found through the start detection threshold and the end detection threshold. However, there is no fixed rule for setting the three thresholds, which relies on manual experience, which places high demands on the user's professional quality. At the same time, the existence of the peak detection threshold will cause some pulses with relatively small amplitudes to be filtered out, and all pulses in the signal cannot be extracted. Gao Leizhen et al. (2019) proposed a double-threshold pulse extraction algorithm. It calculates the average energy and average zero-crossing rate of the sampling points contained in the beginning of a section of background noise in the signal to set the endpoint energy threshold and the endpoint zero-crossing rate threshold, that is, two thresholds. First, the start and end points of the pulse are found through the endpoint energy threshold. Then, the start and end points of the pulse are found again through the endpoint zero-crossing rate threshold. Finally, the final start and end points of the pulse are determined based on the results of the two searches. Similarly, the setting of the endpoint energy threshold and the endpoint zero-crossing rate threshold still relies on manual experience. Although it abandons the use of peak detection thresholds and chooses to directly detect the starting and ending points of the pulse to extract as many pulses as possible, this will also extract the sudden sharp "pulses" from the background noise in the signal, interfering with the quality of the overall pulse. In addition, the average energy and average zero-crossing rate of the beginning of the background noise in the signal can only barely reflect the average level of all background noise, and cannot reflect the average level of the signal. The two thresholds set in this way are not stable. If an accidental background noise or pulse with a large amplitude appears at the beginning of the signal, the two thresholds set will be very large, which will seriously interfere with the pulse extraction results. Therefore, a new pulse extraction algorithm needs to be designed.

[0007] In addition, two acoustic emission sensors are usually used in one-dimensional linear positioning, three acoustic emission sensors are usually used in two-dimensional plane positioning, and four acoustic emission sensors are usually used in three-dimensional spatial positioning. Because the sealed electronic device is a three-dimensional structure, Sun Zhigang and others placed four acoustic emission sensors on its surface to synchronously capture the signal of the superfluous object. In this way, after each round of experiments, the PIND detection equipment will save four segments of superfluous object signals. As mentioned above, every time the superfluous object collides or slides, a pulse will be formed in the superfluous object signal. Because the four acoustic emission sensors are at different distances from the superfluous object, the time at which the pulses are formed in the superfluous object signals they capture is also different, and there is a time difference. In this way, within a fixed experimental time, the superfluous object collides or slides multiple times, and the time of the multiple pulses formed in the four segments of the superfluous object signal is also different. In the time difference positioning method, the attribute information of multiple segments of acoustic emission signals, such as arrival time, energy, etc., is used simultaneously to list equations to calculate the three-dimensional coordinates of the acoustic emission source. Similarly, in the positioning of redundant objects, the four redundant object signals collected in each round of experiments should be combined together for feature extraction, construction of a joint feature vector, and construction of a data set, rather than being four independent individuals for feature extraction, construction of their own feature vectors, and finally merging to construct a data set. This will lose the correlation information between the four acoustic emission sensors. Therefore, it is necessary to match the pulses corresponding to the same collision or sliding of the redundant objects in the four redundant object signals, a total of four pulses, to obtain multiple combinations of four pulses, called pulse groups. In this way, for each pulse group, the characteristic values ​​of the signal features in the multi-domain are calculated on the four pulses it contains, and a feature vector is constructed using all the characteristic values. It effectively retains the correlation information between the four redundant object signals, that is, between the four acoustic emission sensors. Therefore, it is necessary to design an applicable pulse matching algorithm.

[0008] At present, because there are few studies on the positioning of redundant objects, there are also few studies on the pulse extraction algorithms required. There are only two pulse matching algorithms proposed by Sun Zhigang et al. in previous studies. One is the sequential search pulse matching algorithm. It sequentially searches for pulses that appear in sequence in four sections of redundant object signals, and identifies the four pulses searched each time as a pulse group until the last pulse in a certain section of redundant object signal is searched, and the pulse matching ends. Usually, the section of redundant object signal that ends the pulse matching is captured by the acoustic emission sensor farthest from the redundant object. And, precisely because of the farthest distance, if the energy generated by a collision or sliding of the redundant object is small, the acoustic emission sensor may fail to capture the pulse. This will cause a misalignment problem, resulting in the four pulses in some pulse groups not being the same collision or sliding of the source redundant object. The other is the zero-filling time difference pulse matching algorithm. It obtains the time when the first pulse appears in the four-segment redundant signal, takes the redundant signal captured by the acoustic emission sensor farthest from the redundant object as a reference, supplements the background noise of the corresponding time length in the starting stage of the other three redundant signals and discards the redundant signal components in the ending stage, so that the duration of the four redundant signals is still the same. In this way, the appearance time of the pulses in the four redundant signals is matched. On this basis, taking the redundant signal captured by the acoustic emission sensor closest to the redundant object as a reference, each time a pulse appears, the start time and end time of the pulse are obtained, and a signal component is intercepted at the same position in the other three redundant signals, a total of three signal components, which are regarded as three pulses, forming a pulse group. And so on, until the last pulse of the reference redundant signal is reached. It can be seen that the pulse that appears in the redundant signal captured by the acoustic emission sensor closest to the redundant object may not be in the other three redundant signals, especially in the redundant signal captured by the acoustic emission sensor farthest from the redundant object. This will cause individual pulses of some pulse groups to be background noise, and there are missing values ​​or abnormal symbols in the feature vector constructed thereby. At the same time, the reference redundant signal is not stable. The first pulse that appears in the other three redundant signals may not appear in the redundant signal captured by the acoustic emission sensor at the farthest distance. Instead, the first pulse that appears may be a pulse with a larger amplitude, and its corresponding pulse is already behind in the other three redundant signals. This will result in that the four pulses in all pulse groups are not from the same collision or sliding of the source redundant. Summary of the invention

[0009] The present invention aims to solve the problem that the existing multi-channel pulse extraction method has inaccurate pulse extraction for continuous multi-pulses, and the problem that the existing multi-channel pulse extraction and matching method has matching errors caused by missing values ​​or abnormal symbols in some channels.

[0010] A multi-channel pulse extraction method for locating excess objects in sealed electronic equipment, which extracts pulses from each channel signal in a multi-channel signal for locating excess objects in sealed electronic equipment, comprises the following steps:

[0011] Step 1: Calculate the average energy and average zero-crossing rate of the entire signal, expressed as E a and Z a , and the standard deviation of energy and the standard deviation of zero-crossing rate, denoted as σ E and σ Z ; Set the peak detection threshold to E p =E a +3σ E , set the endpoint energy threshold to E be_1 =E a -σ E , set the endpoint zero crossing rate threshold to Z be_1 =Z a -σ Z ;

[0012] Step 2: Frame the signal, set the duration of each frame signal to Δt, and the starting position of the next frame signal to the end position of the previous frame signal; calculate the energy and zero-crossing rate of each frame signal;

[0013] Step 3: Starting from the first frame signal, compare the energy of each frame signal with E p ; When the energy of a frame signal is greater than E p When the frame signal is recorded, the energy of each subsequent frame signal is compared with E p , until the energy of a frame signal is less than E p ; Find the frame signal with the largest energy from these frame signals, and obtain the time of the first sampling point, which is called the peak time t of the current pulse p ;

[0014] Step 4: Take the peak time t of the pulse extracted in step 3 p As the starting point, the energy of each frame signal is compared with E be_1 Compare until the energy of a frame signal is found to be less than E in both directions. be_1 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the first start frame signal and the first end frame signal corresponding to the first level variable multi-threshold pulse of the current pulse;

[0015] The time of the first sampling point in the first start frame signal and the first end frame signal corresponding to the first variable multi-threshold pulse of the current pulse is obtained, which is called the first candidate start time t of the current pulse. b_f1and the first candidate end time t e_f1 ;

[0016] Step 5: Take the peak time t of the pulse extracted in step 3 p As the starting point, the zero-crossing rate of each frame signal is calculated forward and backward respectively. be_1 Compare until the zero-crossing rate of a frame signal is found to be less than Z in both directions. be_1 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the second start frame signal and the second end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse;

[0017] The time of the first sampling point in the second start frame signal and the second end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse is obtained, which is called the second candidate start time t of the current pulse. b_f2 and the second candidate end time t e_f2 ;

[0018] Step 6: Compare the two candidate start times and identify the larger one as the final start time t b_1 Similarly, compare the two candidate end times and identify the smaller one as the final end time t e_1 ;

[0019] Step 7: Take the final end time t of the pulse extracted in step 6 e_1 The next frame signal is taken as a new starting point, and steps 3 to 6 are repeated to extract the second pulse until the last frame signal of the signal is searched; and the extracted multiple pulses are saved;

[0020] Step 8: Reset the endpoint energy threshold to E be_2 =E a +σ E , set the endpoint zero crossing rate threshold to Z be_2 =Z a +σ z ;

[0021] Step 9: Based on the multiple pulses saved in step 7, the peak time t corresponding to the first pulse is p As the starting point, the energy of each frame signal is compared with E be_2 Compare until the energy of a frame signal is found to be less than E in both directions. be_2 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the first start frame signal and the first end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse;

[0022] The time of the first sampling point in the first start frame signal and the first end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse is obtained, which is called the candidate start time t of the current pulse. b_s1 and the candidate end time t e_s1 ;

[0023] Step 10: Take the peak time t of the first pulse saved in step 7 p As the starting point, the zero-crossing rate of each frame signal is calculated forward and backward respectively. be_2 Compare until the zero-crossing rate of a frame signal is found to be less than Z in both directions. be_2 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the second start frame signal and the second end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse;

[0024] The time of the first sampling point in the second start frame signal and the second end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse is obtained, which is called the candidate start time t of the current pulse. b_s2 and the candidate end time t e_s2 ;

[0025] Step 11: Compare the two candidate starting times t b_s1 and t b_s2 , the larger one is identified as the final starting time t b_2 ; Similarly, compare the two candidate end times t e_s1 and t e_s2 , the smaller one is considered as the final ending time t e_2 ;

[0026] Step 12: starting from the second pulse saved in step 7, repeat the processing steps from step 9 to step 11 until all pulses saved in step 7 are processed;

[0027] Step 13: Calculate the first duration = t e_1 -t b_1 and the second duration = t e_2 -t b_2 Calculate the ratio of the second duration to the first duration. If the ratio exceeds the duration ratio threshold, the final starting time t in step six is b_1 and the final ending time t e_1 The real starting time t of the current pulse is considered b and the actual end time t e Otherwise, the final starting time t in step 6 is b_1 and the final end time t in step 11 e_2The real starting time t of the current pulse is considered b and the actual end time t e .

[0028] Preferably, the duration of each frame signal is Δt=50 μs.

[0029] Preferably, the duration ratio threshold is 18%.

[0030] Preferably, the sampling frequency of the multi-pass signal used for locating excess objects in sealed electronic equipment is 500 kHz, and the interval between sampling points is 2 μs.

[0031] A computer storage medium stores at least one instruction, which is loaded and executed by a processor to implement a multi-channel pulse extraction method for locating excess objects in sealed electronic equipment.

[0032] A multi-channel pulse extraction device for locating excess matter in sealed electronic equipment comprises a processor and a memory. The memory stores at least one instruction, which is loaded and executed by the processor to implement a multi-channel pulse extraction method for locating excess matter in sealed electronic equipment.

[0033] A multi-channel pulse matching method for locating excess objects in sealed electronic equipment first uses the multi-channel pulse extraction method for locating excess objects in sealed electronic equipment to extract pulses from each channel signal, and the number of multiple channels is recorded as N; then multi-channel signal matching is performed based on the excess object signal extracted from each channel, and the matching process includes the following steps:

[0034] Step 1: respectively obtain and save the starting time of the pulse in the redundant object signal corresponding to each channel; according to the starting time of the first pulse in the redundant object signal corresponding to each channel, determine the redundant object signal captured by the acoustic emission sensor farthest from the redundant object, and record the number of pulses extracted by the channel corresponding to the sensor as n;

[0035] Step 2: Denote the pulses contained in the excess signal determined in step 1 as P i , i = 1, 2, ..., n, the starting times of these pulses are t i ; Based on the starting time threshold T, the starting time threshold interval of each pulse is obtained [t i -T,t i ];

[0036] Step 3: Starting from P1, search for pulses with a start time within [t1-T, t1] in the redundant signals corresponding to the other N-1 channels; if pulses corresponding to N-1 channels can be found, combine them with p1 to form a pulse group; otherwise, skip p1 if the match fails;

[0037] Step 4: Follow the steps in step 3 to i Pulse matching is performed to obtain multiple pulse groups, thereby completing the pulse matching of the channel redundant signal.

[0038] Preferably, the start time threshold T is 85 μs.

[0039] A computer storage medium stores at least one instruction, which is loaded and executed by a processor to implement a multi-channel pulse matching method for locating excess objects in sealed electronic equipment.

[0040] A multi-channel pulse matching device for locating excess matter in sealed electronic equipment comprises a processor and a memory. The memory stores at least one instruction, which is loaded and executed by the processor to implement a multi-channel pulse matching method for locating excess matter in sealed electronic equipment.

[0041] Beneficial effects:

[0042] The multi-channel pulse extraction method of the present invention can effectively identify each pulse in the continuous multi-pulse, that is, effectively avoid the situation where two pulses are regarded as one pulse, thereby effectively solving the problem of inaccurate pulse extraction in the continuous multi-pulse. At the same time, the multi-channel pulse matching method based on the multi-channel pulse extraction of the present invention can effectively avoid the situation where the farthest acoustic emission sensor may not be able to capture the captured redundant signal, resulting in no pulses, and causing mismatching with corresponding pulses in other channels.

[0043] Therefore, the present invention can not only effectively improve the accuracy of pulse extraction and accurately extract independent pulses from the redundant signal, but also match the pulse group with high correlation from the redundant signal group to improve the matching accuracy. The present invention shifts the focus of the redundant positioning research to the redundant signal for the first time, especially the pulse extraction and pulse matching research. The present invention improves the accuracy of the signal at the source of signal processing, thereby further improving the classification accuracy from the signal source. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of the two-stage variable multi-threshold pulse extraction algorithm.

[0045] Figure 2 Schematic diagram of continuous multiple pulses.

[0046] Figure 3 Schematic diagram of the pulse matching algorithm for threshold decision search.

[0047] Figure 4 This is a model diagram of aerospace power supply.

[0048] Figure 5 This is the processing effect of the two-level variable multi-threshold pulse extraction algorithm, where from top to bottom are the redundant object signals captured by acoustic emission sensors No. 1 to No. 4.

[0049] Figure 6 The processing effect of searching the pulse matching algorithm for threshold determination.

[0050] Figure 7 Classification accuracy achieved by random forest and support vector machine on five datasets.

[0051] Figure 8 Precision, Recall, and F1 scores obtained by random forest on the positioning dataset.

[0052] Fig. 9 Precision, Recall and F1 scores obtained by the support vector machine on the positioning dataset. DETAILED DESCRIPTION

[0053] In view of the shortcomings in the existing pulse extraction algorithm and pulse matching algorithm research, as well as the latest demand for redundant object detection, the present invention proposes a four-channel pulse extraction and matching algorithm suitable for redundant object positioning, in order to further improve the classification accuracy achieved by the classifier. First, the present invention proposes a two-stage variable multi-threshold pulse extraction algorithm. The algorithm includes two stages. In the first stage, it calculates the average energy and average zero-crossing rate of all sampling points in the signal, and sets the peak energy threshold, endpoint energy threshold and endpoint zero-crossing rate threshold according to the 3σ distribution criterion of the normal distribution. The highest point of the pulse is first quickly found through the peak energy threshold, the starting point and end point of the pulse are found through the endpoint energy threshold, and then the starting point and end point of the pulse are found again through the endpoint zero-crossing rate threshold. The starting point and end point of the pulse in the first stage are determined according to the results of the two searches. In the second stage, the endpoint energy threshold and the endpoint zero-crossing rate threshold are reset again according to the 3v distribution criterion of the normal distribution, and the starting point and end point of the two pulses are found through the same process, and the starting point and end point of the second stage pulse are determined. This effectively solves the problems existing in the existing research, such as the threshold setting rules are not fixed, the threshold reference basis is not unified, the pulse extraction reliability is not high, and continuous multiple pulses are not considered. Secondly, the present invention proposes a threshold judgment search pulse matching algorithm. It is based on a two-level variable multi-threshold pulse extraction algorithm. It first searches for the start time of the first pulse in the four-segment redundant signal. Taking the start time of each pulse in the redundant signal captured by the acoustic emission sensor farthest from the redundant object as a reference, the pulse that meets the start time threshold requirement is searched in the other three redundant signals. For each pulse, if a pulse that meets the requirements can be found in the other three redundant signals, they are grouped into a pulse group. As long as a pulse that meets the requirements cannot be found in a redundant signal, the current pulse is skipped and not matched. This effectively solves the problems existing in the existing research, such as unreasonable reference object setting and unstable and unreliable pulse matching. The following is an explanation in conjunction with a specific implementation method. Specific implementation method one:

[0055] This embodiment is a multi-channel pulse extraction and matching method for positioning excess objects in sealed electronic equipment. Four acoustic emission sensors can be used to complete the processing in three-dimensional space positioning. Therefore, this embodiment uses four-channel pulses corresponding to four acoustic emission sensors for description, that is, this embodiment is a four-channel pulse extraction and matching method for positioning excess objects in sealed electronic equipment, including the following steps:

[0056] S1, two-level variable multi-threshold pulse extraction algorithm:

[0057] According to the 3σ distribution criterion of the normal distribution, for a batch of values, assuming that their mean is μ and their standard deviation is σ, then the probability that the value is distributed in [μ-σ, μ+σ] is 68.27%, and the probability that the value is distributed in [μ-3σ, μ+3σ] is 99.73%. On this basis, scholars usually regard values ​​lower than μ-3σ or higher than μ+3σ as abnormal values, among which μ+3σ is an abnormally large value, and regard values ​​hovering around μ-σ and μ+σ as valid values. In other words, for a signal, when calculating its average energy Z a , average zero-crossing rate e a , the standard deviation of energy σ E and the standard deviation of the zero-crossing rate σ z Then, we can a +3σ E Find the highest point of the pulse, i.e. the peak value, in the signal component of a -σ E and E a +σ E Near Z a -σ Z and Z a +σ Z Find the start and end points of the pulse in nearby signal components. Figure 1 A schematic diagram of a two-stage variable multi-threshold pulse extraction algorithm is given, and its specific implementation steps are as follows:

[0058] Step 1: Calculate the average energy and average zero-crossing rate of the entire signal, expressed as E a and Z a , and the standard deviation of energy and the standard deviation of zero-crossing rate, denoted as σ E and σ z On this basis, the peak detection threshold is set to E p =E a +3σ E , set the endpoint energy threshold to E be_1 =E a -σ E , set the endpoint zero crossing rate threshold to Z be_1 =Z a -σ z .

[0059] Step 2: Frame the signal and set the duration of each frame signal to Δt = 50μs. In fact, it is to use Δt as a window sliding frame, and the starting position of the next frame signal is the ending position of the previous frame signal; calculate the energy and zero-crossing rate of each frame signal.

[0060] Since the sampling frequency of the PIND detection device used in this embodiment is 500 kHz, the interval between sampling points is 2 μs, and a 50 μs frame signal contains 25 sampling points.

[0061] Step 3: Starting from the first frame signal, compare the energy of each frame signal with E p When the energy of a frame signal is greater than E p When the frame signal is recorded, the energy of each subsequent frame signal is compared with E p , until the energy of a frame signal is less than E p Find the frame signal with the largest energy from these frame signals and obtain the time of the first sampling point, which is called the peak time t of the current pulse. p .

[0062] Step 4: Take the peak time t of the pulse extracted in step 3 o As the starting point, the energy of each frame signal is compared with E be_1 Compare until the energy of a frame signal is found to be less than E in both directions. be_1 The previous frame signals of the two frame signals are respectively identified as the start frame signal and the end frame signal of the current pulse, and recorded as the first start frame signal and the first end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse.

[0063] The time of the first sampling point in the first start frame signal and the first end frame signal corresponding to the first variable multi-threshold pulse of the current pulse is obtained, which is called the first candidate start time t of the current pulse. b_f1 and the first candidate end time t e_f1 .

[0064] Step 5: Take the peak time t of the pulse extracted in step 3 p As the starting point, the zero-crossing rate of each frame signal is calculated forward and backward respectively. be_1 Compare until the zero-crossing rate of a frame signal is found to be less than Z in both directions. be_1 The previous frame signals of the two frame signals are respectively identified as the start frame signal and the end frame signal of the current pulse, and recorded as the second start frame signal and the second end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse.

[0065] The time of the first sampling point in the second start frame signal and the second end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse is obtained, which is called the second candidate start time t of the current pulse. b_f2 and the second candidate end time t e_f2 .

[0066] Step 6: Compare the two candidate start times and identify the larger one as the final start time t b_1 Similarly, compare the two candidate end times and identify the smaller one as the final end time t e_1 .

[0067] Step 7: Take the final end time t of the pulse extracted in step 6 e_1 The next frame signal is taken as the new starting point, and steps 3 to 6 are repeated to extract the second pulse. This process is repeated until the last frame signal of the signal is found. The extracted multiple pulses are saved.

[0068] At this point, the first stage of processing is completed, that is, the first-level variable multi-threshold pulse extraction algorithm is completed.

[0069] Step 8: Reset the endpoint energy threshold to E be_2 =E a +σ E , set the endpoint zero crossing rate threshold to Z be_2 =Z a +σ z .

[0070] Step 9: Based on the multiple pulses saved in step 7, the peak time t corresponding to the first pulse is p As the starting point, the energy of each frame signal is compared with E be_2 Compare until the energy of a frame signal is found to be less than E in both directions. be_2 The previous frame signals of the two frame signals are respectively identified as the start frame signal and the end frame signal of the current pulse, and recorded as the first start frame signal and the first end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse.

[0071] The time of the first sampling point in the first start frame signal and the first end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse is obtained, which is called the candidate start time t of the current pulse. b_s1 and the candidate end time t e_s1 .

[0072] Step 10: Take the peak time t of the first pulse saved in step 7 p As the starting point, the zero-crossing rate of each frame signal is calculated forward and backward respectively. be_2 Compare until the zero-crossing rate of a frame signal is found to be less than Z in both directions. be_2 The previous frame signals of the two frame signals are respectively identified as the start frame signal and the end frame signal of the current pulse, and recorded as the second start frame signal and the second end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse.

[0073] The time of the first sampling point in the second start frame signal and the second end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse is obtained, which is called the candidate start time t of the current pulse. b_s2 and the candidate end time t e_s2 .

[0074] Step 11: Compare the two candidate starting times t b_s1 and t b_s2 , the larger one is identified as the final starting time t b_2 Similarly, compare the two candidate end times t e_s1 and t e_s2 , the smaller one is considered as the final ending time t e_2 .

[0075] Step 12: Starting from the second pulse saved in step 7, repeat the processing steps from step 9 to step 11. And so on, until all the pulses saved in step 7 are processed.

[0076] Step 13: Calculate the first duration = t e_1 -t b_1 and the second duration = t e_2 -t b_2 , that is, the difference between the end time and the start time of each pulse in step 6 and the difference between the end time and the start time in step 11. Calculate the ratio of the second duration to the first duration. If the ratio exceeds 18%, the final start time t in step 6 is set to b_1 and the final ending time t e_1 The real starting time t of the current pulse is considered b and the actual end time t e If the ratio is lower than 18%, the final starting time t in step 6 is b_1 and the final end time t in step 11 e_2 The real starting time t of the current pulse is considered b and the actual end time t e .

[0077] Step 14: The pulse processed in step 13 is retained, and the pulse extraction of the entire signal is completed.

[0078] At this point, the second stage of processing is completed, that is, the two-stage variable multi-threshold pulse extraction algorithm is completed.

[0079] In fact, step thirteen is to solve the problem of continuous multiple pulses, which is also not considered by the three-threshold pulse extraction algorithm and the double-threshold pulse extraction algorithm. Figure 2A schematic diagram of continuous multiple pulses is given. In the three-threshold pulse extraction algorithm, after the peak detection threshold is used to find the maximum frame signal of a pulse, the end frame signal found by the end detection threshold may be the end frame signal of the next pulse, not the true end frame signal of the current pulse, as shown in the box in the figure. The reason is that the signal attenuation between the two pulses does not reach the set end detection threshold. Similarly, in the double-threshold pulse extraction algorithm, the endpoint energy threshold plays the same role as the end detection threshold of the three-threshold pulse extraction algorithm, and the difference between them is not large. Therefore, when the endpoint energy threshold is used for pulse extraction, it will also regard the two pulses in the box as one pulse. When a comprehensive analysis is performed on the pulse extraction results obtained using the endpoint energy threshold and the endpoint zero-crossing rate threshold, if the endpoint zero-crossing rate threshold plays a dominant role, the problem of continuous multiple pulses may be avoided. But when the endpoint energy threshold plays a dominant role, the problem of continuous multiple pulses is inevitable. Therefore, it is emphasized in the present invention. It should be noted that for the purpose of visual display, Figure 2 What is displayed is a continuous analog signal, not a discrete digital signal.

[0080] In the two-stage variable multi-threshold pulse extraction algorithm, the endpoint energy threshold E is set for the first time. be_1 Or endpoint zero crossing rate threshold Z be_1 It is higher than the end detection threshold of the three-threshold pulse extraction algorithm and the endpoint energy threshold or endpoint zero-crossing rate threshold of the double-threshold pulse extraction algorithm, because it filters the signal components with energy lower than μ-σ in the signal, and the filtering ratio is higher than 15.73%. This has raised the entry threshold of the frame signal belonging to the pulse to a certain extent. In addition, the endpoint energy threshold E be_2 Or endpoint zero crossing rate threshold Z be_2 The signal components with energy lower than μ+σ in the signal are filtered out, and the filtering ratio is higher than 84%. Therefore, if E is used directly be_2 or Z be_2 When performing pulse extraction, many pulses will be discarded, resulting in a small number of pulses extracted. be_2 or Z be_2 It is only used to judge continuous multiple pulses. For frame signals belonging to independent pulses, use E be_2 or Z be_2 The extracted signal components are compared with the be_1 or Z be_1 The ratio of the extracted signal components should be approximately equal to That is, the ratio of durations. When it comes to continuous multiple pulses, its distribution is a combination of the distributions of multiple independent pulses, so the ratio of durations should be less than 18.73%. Therefore, the present invention sets 18% as the final comparison threshold, which retains a moderate elastic space and allows the occurrence of partial degradation phenomena in actual situations. For independent pulses, the ratio of durations is greater than 18%, and its starting point and end point refer to the processing results of the first stage. For continuous multiple pulses, the ratio of durations is less than 18%, and its starting point and end point refer to the processing results of the first stage and the second stage respectively, and take their best union, such as Figure 2 In fact, the result of processing multiple consecutive pulses is to effectively extract the first pulse among them.

[0081] S2, threshold judgment search pulse matching algorithm:

[0082] Pulse matching is completed on the basis of pulse extraction. That is, firstly, a two-stage variable multi-threshold pulse extraction algorithm is used to process the four sections of redundant signals respectively, and the extracted pulses are retained respectively, and then a threshold judgment search pulse matching algorithm is used to determine the pulse group of the four sections of redundant signals. Figure 3 A schematic diagram of the threshold judgment search pulse matching algorithm is given, and its specific implementation steps are as follows:

[0083] Step 1: Obtain and save the start time of the pulses in the four sections of the redundant object signal respectively. According to the start time of the first pulse in the four sections of the redundant object signal, determine the redundant object signal captured by the acoustic emission sensor farthest from the redundant object.

[0084] Step 2: The pulses contained in the redundant signal determined in step 1 are represented as P i (i=1,2,…,n), the starting times of these pulses are t i (i=1,2,…,n). Determine the starting time threshold T and obtain the starting time threshold interval [t i -T,t i ](i=1,2,…,n).

[0085] Step 3: Starting from P1, search for pulses with a start time within [t1-T, t1] in the other three redundant signals. If three pulses can be found, combine them with P1 to form a pulse group. Otherwise, skip P1, indicating that its matching is unsuccessful.

[0086] Step 4: Refer to step 3 and i (i=2,3,…,n) to perform pulse matching. At this point, the pulse matching of four sections of redundant signals is completed, and multiple pulse groups are obtained.

[0087] It should be noted that the start time of the first pulse in the superfluous object signal captured by the acoustic emission sensor farthest from the superfluous object must be the longest. If the energy generated by the collision or sliding of the superfluous object is large, then all four acoustic emission sensors can capture this energy, and the acoustic emission sensor farthest away will be the last to capture it, that is, the start time of the pulse is the longest. If the energy generated by the collision or sliding of the superfluous object is small, then only the closest few acoustic emission sensors can capture this energy, and the acoustic emission sensor farthest away may not be able to capture it, so there is no pulse in the superfluous object signal it captures. It may not be able to capture it until the next time the superfluous object collides or slides to generate a large energy, and then generate the first pulse in the superfluous object signal. In this way, the start time of the first pulse in the superfluous object signal captured by the acoustic emission sensor farthest from the superfluous object is even longer. Therefore, when taking the pulse in the superfluous object signal captured by the acoustic emission sensor farthest from the superfluous object as a reference, it is necessary to set the start time threshold interval [t i -T,t i ] to find the pulses that meet the requirements in the other three redundant signals, rather than in the starting time threshold interval [t i ,t i +T]. In addition, the determination of the starting time threshold T needs to refer to the time difference between the pulses with the largest amplitude that are obviously correlated in the four sections of the redundant signal. The specific analysis process will be given later.

[0088] Thus, the description of the four-channel pulse extraction and matching algorithm proposed by the present invention ends. It specifically includes two stages: a two-stage variable multi-threshold pulse extraction algorithm and a threshold judgment search pulse matching algorithm, which is a sequential processing relationship. The four-channel pulse extraction and matching algorithm mentioned in the embodiment is used for processing, that is, the two-stage variable multi-threshold pulse extraction algorithm is used to process all the redundant signals first, and then the threshold judgment search pulse matching algorithm is used to process the four-segment redundant signals. Specific implementation method 2:

[0090] This embodiment is a computer storage medium, wherein at least one instruction is stored in the storage medium, wherein the at least one instruction at least includes a program instruction corresponding to a multi-channel pulse extraction method for locating excess objects in a sealed electronic device, which may be a program instruction corresponding to a multi-channel pulse extraction method for locating excess objects in a sealed electronic device, or may be a program instruction corresponding to a multi-channel pulse matching method for locating excess objects in a sealed electronic device that also includes a matching process. The at least one instruction is loaded and executed by a processor to implement the multi-channel pulse extraction method for locating excess objects in a sealed electronic device / multi-channel pulse matching method for locating excess objects in a sealed electronic device.

[0091] It should be understood that the instructions include computer program products, software or computerized methods corresponding to any method described in the present invention; the instructions can be used to program a computer system, or other electronic device; computer storage media may include readable media on which instructions are stored, which may include but are not limited to magnetic storage media, optical storage media; magneto-optical storage media include read-only memory ROM, random access memory RAM, erasable programmable memory (for example, EPROM and EEPROM) and flash memory layers, or other types of media suitable for storing electronic instructions. Specific implementation method three:

[0093] This embodiment is a multi-channel pulse extraction device for locating excess materials in sealed electronic equipment. The device includes a processor and a memory. It should be understood that the device includes any device including a processor and a memory described in the present invention. The device may also include other units and modules that perform display, interaction, processing, control, etc. and other functions through signals or instructions.

[0094] At least one instruction is stored in the memory, and the at least one instruction at least includes a program instruction corresponding to a multi-channel pulse extraction method for locating excess objects in a sealed electronic device, which may be a program instruction corresponding to a multi-channel pulse extraction method for locating excess objects in a sealed electronic device, or may be a program instruction corresponding to a multi-channel pulse matching method for locating excess objects in a sealed electronic device that also includes a matching process. The at least one instruction is loaded and executed by the processor to implement the multi-channel pulse extraction method for locating excess objects in a sealed electronic device / multi-channel pulse matching method for locating excess objects in a sealed electronic device.

[0095] Example

[0096] 1Signal Source

[0097] A certain type of cubic aerospace power supply is selected as the sealed electronic device of the present invention, and its interior is composed of a large number of battery rows. There is a confined space at each of the eight vertices of the aerospace power supply, for a total of eight confined spaces. There is a high probability that excess matter will be generated in these eight confined spaces and will be in a state of random motion, generating excess matter signals. Therefore, for the location of excess matter in this type of aerospace power supply, the classifier needs to solve the eight-classification problem. Four acoustic emission sensors are placed at different positions on the surface of the aerospace power supply, all of which are connected to the PIND detection equipment, and their signal sensitivity capture range covers the entire aerospace power supply. Figure 4 A model diagram of an aerospace power supply is given. Its interior is abstractly divided into eight confined spaces, each of which is named. At the same time, the placement of four acoustic emission sensors is shown.

[0098] The newly detected excess material from aerospace power supply in real application scenarios, namely 0.2 mg of solder particles, was selected and placed in a confined space numbered 1. Mechanical excitation was generated by the vibration table, so that the excess material was in a random motion state and an excess material signal was generated. In this process, the PIND detection equipment collected and saved a large number of excess material signals, which were labeled "1", indicating that they were generated in the confined space numbered 1. Similarly, the excess material was placed in confined spaces numbered 2 to 8, and a large number of excess material signals were collected and saved using the vibration table and PIND detection equipment. They were labeled "2" to "8", indicating that they were generated in the confined spaces numbered 2 to 8. Finally, 100 groups of excess material signals labeled "1" to "8" were screened, for a total of 800 groups of excess material signals. Each group of excess material signals contains four segments of excess material signals.

[0099] 2 Visualization

[0100] The four-channel pulse extraction and matching algorithm was used to process 800 groups of redundant signals, and the number of pulses in each group of redundant signals after processing, as well as the start and end time of each pulse, were retained. In order to visualize the four-channel pulse extraction and matching algorithm, one group of redundant signals was randomly selected, and the processing effect was displayed in two stages. The first stage was to use a two-level variable multi-threshold pulse extraction algorithm for processing. Figure 5 The processing effect of the first stage is shown, and from top to bottom, the redundant signals captured by acoustic emission sensors No. 1 to No. 4 are shown. The red short vertical line in the figure indicates the start time of the pulse, and the blue short vertical line indicates the end time of the pulse.

[0101] As can be seen from the figure, the pulses in the four sections of the superfluous signal are effectively extracted. Among them, there is a continuous pulse in one interval in the first half and the second half of them, and the two intervals of each section of the superfluous signal are enlarged and displayed. It can be seen that although these pulses are very close, the two-level variable multi-threshold pulse extraction algorithm still extracts them accurately and independently, and there is no problem of continuous multiple pulses. In addition, there are some pulses with relatively small amplitudes in the four sections of the superfluous signal that are not extracted. On the one hand, they are identified as sudden changes in the background noise. On the other hand, they are identified as pulses with amplitudes that are too small to be worth extracting. Overall, almost all pulses in the four sections of the superfluous signal are effectively extracted. This effectively illustrates the feasibility and superiority of the two-level variable multi-threshold pulse extraction algorithm proposed by the present invention. According to statistics, 275 pulses are extracted from the first section of the superfluous signal, that is, the superfluous signal captured by the No. 1 acoustic emission sensor, and 341, 344 and 315 pulses are extracted from the superfluous signals captured by the No. 2, No. 3 and No. 4 acoustic emission sensors, respectively. From this, we can preliminarily analyze that the No. 3 acoustic emission sensor is closest to the redundant object, and the No. 1 acoustic emission sensor is farthest from the redundant object. Figure 5 It can also be seen in (a) that some pulses with relatively small amplitudes in the signal of the superfluous object captured by acoustic emission sensor No. 1 are not extracted. This may be because the energy generated by the collision or sliding of the superfluous object is small, so it is already very weak when it propagates to acoustic emission sensor No. 1. On this basis, the second stage is to use the threshold judgment search pulse matching algorithm for processing. Figure 6 The processing effect of the second stage is given.

[0102] As can be seen from the figure, the positions of the matched pulses in the four sections of the superfluous signal are close, and the number of matched pulses is the same. Obviously, the pulses in the superfluous signal captured by acoustic emission sensor No. 1 are referenced in the pulse matching process, because after pulse matching, the pulses in the superfluous signal captured by acoustic emission sensor No. 1 are almost all retained, and the pulses in the other three sections of the superfluous signal are discarded. According to statistics, after pulse matching, there are only 266 pulses left in the four sections of the superfluous signal. Compared with before pulse matching, only 9 pulses in the superfluous signal captured by acoustic emission sensor No. 1 were discarded, and the other 266 pulses found "members" in the other three sections of the superfluous signal. This effectively illustrates the feasibility and superiority of the threshold judgment search pulse matching algorithm proposed in the present invention.

[0103] It should be noted that, for the 800 groups of superfluous signals used in the present invention, after analysis, it is determined that the starting time threshold T is 85μs. Specifically, a two-level variable multi-threshold pulse extraction algorithm is used to process each group of superfluous signals, and the largest frame signal among the four segments of superfluous signals in each group is obtained, and their time is recorded. The difference between the maximum time and the minimum time is calculated. The time difference of about 800 superfluous signals is counted, and the 800 time differences are analyzed to obtain their upper limits. It is concluded that the maximum time difference is 80.31μs. On this basis, the starting time threshold T is set to 85μs. It not only includes the maximum time difference, but also reserves a certain amount of elastic space to allow mutations to occur. Multiple experiments have proved that this setting is reasonable and stable.

[0104] 3-stage evaluation effect

[0105] The two stages of the four-channel pulse extraction and matching algorithm proposed in the present invention are evaluated separately, using the two indicators of pulse number and pulse utilization rate respectively. First, the two-stage variable multi-threshold pulse extraction algorithm of the first stage is evaluated. 800 groups of redundant signals are processed using the existing three-threshold pulse extraction algorithm and the double-threshold pulse extraction algorithm, respectively, and the number of pulses of each group of redundant signals after processing is retained. A representative group of redundant signals is randomly selected from 100 groups of redundant signals labeled "1" to "8", a total of 8 groups and 32 segments of redundant signals, and the number of pulses extracted from them by the three pulse extraction algorithms is counted, as shown in Table 1.

[0106] Table 1 Number of pulses extracted by three pulse extraction algorithms

[0107]

[0108]

[0109] It can be seen from the table that no matter which group or segment of the redundant signal, the double threshold pulse extraction algorithm extracts the most pulses. As described in the introduction, the double threshold pulse extraction algorithm directly uses the endpoint energy threshold and the endpoint zero crossing rate threshold to find the starting point and the end point of the pulse, and lacks the process of using the peak detection threshold to pre-screen the pulse, so it will extract all the pulses in the signal as much as possible, including the background noise from the mutation and the pulse with a smaller amplitude. Like the three-threshold pulse extraction algorithm, the two-level variable multi-threshold pulse extraction algorithm proposed in the present invention also uses the peak detection threshold for pulse pre-screening. Therefore, the number of pulses extracted by these two pulse extraction algorithms is less than that of the double threshold pulse extraction algorithm. It is worth noting that the number of pulses extracted by the two-level variable multi-threshold pulse extraction algorithm proposed in the present invention is more than that of the three-threshold pulse extraction algorithm. On the one hand, this effectively illustrates the advantage of the two-level variable multi-threshold pulse extraction algorithm proposed in the present invention in accurately extracting pulses, that is, finding as many pulses as possible. On the other hand, this also effectively illustrates the advantage of the two-level variable multi-threshold pulse extraction algorithm proposed in the present invention in dealing with the problem of continuous multiple pulses, that is, accurately extracting independent pulses. In general, the number of pulses extracted by the two-stage variable multi-threshold pulse extraction algorithm proposed in the present invention is relatively appropriate and stable, which can illustrate its feasibility and superiority. In addition, from the number of pulses extracted by the three pulse extraction algorithms, it is indeed that the acoustic emission sensor No. 4 is closest to the redundant object, and the acoustic emission sensor No. 1 is farthest from the redundant object. This is consistent with the analysis results in 2.

[0110] Secondly, the threshold judgment search pulse matching algorithm of the second stage is evaluated. Referring to the control variable method, 800 groups of superfluous signals processed by the same pulse extraction algorithm are selected, and on this basis, the existing sequential search pulse matching algorithm and the zero-filling time difference pulse matching algorithm are used for processing, and the number of pulses of each group of superfluous signals after processing is retained, that is, the same number of pulses after pulse matching. In order to facilitate the comparison with the threshold judgment search pulse matching algorithm proposed in the present invention, 800 groups of superfluous signals processed by the two-stage variable multi-threshold pulse extraction algorithm are selected. After the 800 groups of superfluous signals are processed using the three pulse matching algorithms, the pulse utilization rate of each group of superfluous signals is calculated, that is, the ratio of the number of pulses retained by each segment of superfluous signal after pulse matching to the number of pulses extracted before pulse matching (i.e., after pulse extraction). Similarly, 8 groups of superfluous signals and a total of 32 segments of superfluous signals displayed in the pulse extraction stage are selected, and the pulse utilization rate of each group of superfluous signals after processing by the three pulse matching algorithms is statistically analyzed, as shown in Table 2. It should be noted that when the zero-filling time difference pulse matching algorithm is used for processing, it will add background noise at the beginning stage of the three segments of the redundant signal except for the redundant signal captured by the acoustic emission sensor farthest from the redundant object, and discard the redundant signal components at the end stage. In fact, the pulses that are not in the redundant signal captured by the acoustic emission sensor farthest from the redundant object are discarded. Therefore, for the zero-filling time difference pulse matching algorithm, the number of pulses of the four segments of the redundant signal after it is processed is a value between the minimum number of pulses and the maximum number of pulses before processing.

[0111] Table 2 Pulse utilization of redundant signals after processing by three pulse matching algorithms

[0112]

[0113]

[0114] It can be seen from the table that the pulse utilization rate of the redundant signal after the threshold judgment search pulse matching algorithm proposed in the present invention is the lowest. This is because it uses the redundant signal with the least number of pulses in the redundant signal group as a reference, and searches for "group members" that meet the requirements in the other three redundant signals. The number of pulses retained after pulse matching must be less than or equal to the number of pulses in the reference redundant signal. Therefore, the calculated pulse utilization rate is relatively low. Figure 2Taking the four segments of redundant signals shown as an example, the number of pulses retained after pulse matching is 266, which is lower than the 275, 341, 344 and 315 pulses before pulse matching. On the contrary, although the sequential search pulse matching algorithm essentially uses the redundant signal with the least number of pulses in the redundant signal group as a reference, it directly arranges the "group members" in the other three segments of redundant signals in order without making a "group member rationality" judgment. This leads to the problem of unstable and unreliable pulse matching. At this time, the number of pulses retained after pulse matching is the same as the number of pulses in the reference redundant signal. Figure 2 Taking the four segments of redundant signals shown as an example, the number of pulses retained after pulse matching is 275, which is consistent with the minimum number of pulses before pulse matching. Compared with the threshold judgment search pulse matching algorithm, in the process of calculating the pulse utilization rate, although the denominator used for calculation is the same, the numerator used for calculation by the sequential search pulse matching algorithm is significantly larger. This explains why the pulse utilization rate of the redundant signal after being processed by the threshold judgment search pulse matching algorithm is the lowest. The pulse utilization rate of the redundant signal after being processed by the zero-filled time difference pulse matching algorithm is the highest, and most of them are close to or exceed 100%. This is because it uses the redundant signal captured by the acoustic emission sensor closest to the redundant object that discards some signal components as a reference, and searches for or creates "members" in the other three segments of redundant signals. Judging from the pulse utilization rates of the other three segments of redundant signals, it is obvious that it regards some signal components from the background noise as pulses. Figure 2 Taking the four segments of redundant signals shown as an example, the number of pulses retained after pulse matching is 323, which is a value between 275 and 344.

[0115] Comprehensively considered, whether using the number of pulses to evaluate the three pulse extraction algorithms or using the pulse utilization rate to evaluate the three pulse matching algorithms, it is impossible to give an intuitive and reliable evaluation result that feeds back their performance. The evaluation results given by using these two indicators can better feed back the characteristics of each algorithm itself, especially the algorithm principle. This shows that evaluating the four-channel pulse extraction and matching algorithm proposed in the present invention in stages cannot produce satisfactory evaluation results. Therefore, it is necessary to use intuitive and reliable indicators to evaluate the feasibility and superiority of the four-channel pulse matching algorithm proposed in the present invention as a whole.

[0116] 4 Overall evaluation effect

[0117] As described in the background technology, in the positioning of redundant objects, if the classification performance of the classifier is improved from the source, it is necessary to accurately extract high-quality pulses from the redundant object signals. In addition, four acoustic emission sensors are used for redundant object positioning, so it is necessary to combine the four sections of redundant object signals in the process of constructing the data set, and it is necessary to accurately match the relevant pulse groups from the four sections of redundant object signals. Therefore, for 800 groups of redundant object signals, a four-channel pulse extraction and matching algorithm is used to process them, and the characteristic values ​​of the signal features in the multi-domain are calculated on each pulse group to construct a feature vector. According to the label of the redundant object signal group to which the pulse group belongs, a label is added to the feature vector to obtain labeled data, thereby constructing a data set, which is called a positioning data set. Previous studies have shown that extracting the 34 signal features shown in Table 3 from the time domain, frequency domain and Mel frequency cepstral coefficients can feedback the differences in acoustic characteristics between the redundant object signals generated at different locations. Therefore, in the present invention, the characteristic values ​​of these signal features are calculated on each pulse group, and each labeled data obtained thereby is 34×4+1=137 dimensions.

[0118] Table 3 34 signal characteristics

[0119]

[0120]

[0121] At the same time, in order to compare with the four-channel pulse extraction and matching algorithm, the four combined algorithms of "three-threshold pulse extraction algorithm + sequential search pulse matching algorithm", "three-threshold pulse extraction algorithm + zero-time difference pulse matching algorithm", "double-threshold pulse extraction algorithm + sequential search pulse matching algorithm" and "double-threshold pulse extraction algorithm + zero-time difference pulse matching algorithm" were used to process 800 groups of redundant signals respectively, and the pulse groups belonging to the four combined algorithms were obtained respectively. Similarly, the characteristic values ​​of 34 signal features were calculated on each pulse group, the characteristic vector was constructed and the label was added to obtain a large amount of labeled data. According to the combined algorithm used corresponding to the pulse group, the large amount of labeled data was divided into four parts to form four data sets, which were named experimental data set 1, experimental data set 2, experimental data set 3 and experimental data set 4, respectively corresponding to the combined algorithms of "three-threshold pulse extraction algorithm + sequential search pulse matching algorithm", "three-threshold pulse extraction algorithm + zero-time difference pulse matching algorithm", "double-threshold pulse extraction algorithm + sequential search pulse matching algorithm" and "double-threshold pulse extraction algorithm + zero-time difference pulse matching algorithm". Table 4 gives the specific description of the five data sets.

[0122] Table 4 Detailed description of the five datasets

[0123]

[0124] Most classifiers in machine learning are based on tree theory or mathematical theory. Commonly used classifiers based on tree theory include decision trees, random forests, XGBoost, etc. Commonly used classifiers based on mathematical theory include naive Bayes, k-nearest neighbor, support vector machine, etc. In order to comprehensively evaluate the quality of the five data sets, and then evaluate the processing effects of the four-channel pulse extraction and pulse matching algorithm proposed in the present invention and the other four combined algorithms, representative classifiers based on tree theory and mathematical theory, random forests and support vector machines, were selected respectively. Considering that all five data sets contain data with 8 kinds of labels, radial basis functions are introduced for support vector machines. The five data sets are divided into five training sets and five test sets in a ratio of 7:3. Two classifiers are trained on the training sets respectively, and their classification performance is tested on the corresponding test sets, and the classification accuracy obtained is statistically calculated. Figure 7 It should be noted that the two classifiers trained here are configured with default parameters and no parameter optimization is performed.

[0125] from Figure 7 It can be clearly seen that both classifiers achieve the highest classification accuracy on the positioning data set, and the advantages are obvious. This effectively illustrates the superiority of the four-channel pulse extraction and matching algorithm proposed in the present invention. At the same time, compared with the classification accuracy of more than 95% achieved by the existing redundant object positioning research mentioned in the introduction, the classification accuracies achieved by the two classifiers without parameter optimization are 98.27% and 97.64%, respectively, which are significantly higher than those achieved in the existing research, not to mention the parameter optimization of the two classifiers. This effectively illustrates the correctness and necessity of carrying out the present invention. In addition, the classification accuracies achieved by the two classifiers on experimental data set 1 and experimental data set 2 are higher than those achieved on experimental data set 3 and experimental data set 4, respectively. This shows that the processing effect of the three-threshold pulse extraction algorithm is better than that of the double-threshold pulse extraction algorithm, and it can extract pulses with higher reliability from the redundant object signal, rather than more pulses. This once again illustrates the importance of pulse pre-screening from the side. The classification accuracies achieved by the two classifiers on experimental data set 1 and experimental data set 3 are slightly higher than those achieved on experimental data set 2 and experimental data set 4, respectively. This shows that the processing effect of the sequential search pulse matching algorithm is better than that of the zero-filled time difference pulse matching algorithm. In fact, it can be seen from the previous description that the processing effects of these two pulse matching algorithms should be average. However, in the present invention, due to the effect of the pulse extraction algorithm, the reliability of the correlation between the pulses that appear sequentially in the four sections of the redundant signal group is greatly improved. In this way, the effect of the sequential search pulse matching algorithm is brought into play. The zero-filled time difference pulse matching algorithm creates some pulses from the background noise, causing some dirty data to enter the experimental data set, resulting in a decrease in the classification accuracy achieved by the classifier.

[0126] In addition, the embodiment calculates the Precision, Recall and F1 scores of the two classifiers on the positioning dataset, and observes their classification performance on the data of each label, such as Figure 8-Figure 9 shown.

[0127] from Figure 8-Figure 9 It can be seen that both classifiers achieve similar Precision, Recall and F1 scores on the data of 8 labels in the positioning data set. This shows that the classification performance of the two classifiers is stable and reliable. In fact, Precision, Recall and F1 scores depend not only on the overall quality of the data of each label, but also on the amount of data of each label. From this perspective, the four-channel pulse extraction and matching algorithm proposed in the present invention can extract pulses with high reliability and match pulse groups with large correlation as much as possible, thereby obtaining high-quality data, and can stably process the redundant signal groups generated in each confined space, thereby ensuring data balance. This fully demonstrates the feasibility and superiority of the four-channel pulse extraction and matching algorithm proposed in the present invention.

[0128] During the production process of sealed electronic devices, redundant objects may be introduced, which may affect their reliable operation. The PIND method is a commonly used method for detecting redundant objects. For sealed electronic devices with large weight and volume and complex internal structure, obtaining the location of redundant objects can improve the efficiency of the redundant object cleaning work. At present, the research on redundant object positioning has been carried out in the aspects of classifier selection and optimization, data set processing and optimization, and multi-domain signal feature extraction, but no attention has been paid to the redundant object signals at the source, especially pulse extraction. In addition, for sealed electronic devices with three-dimensional structures, the four-channel redundant object signals need to be combined to construct feature vectors to retain the correlation between them. Therefore, a suitable pulse matching algorithm needs to be designed. However, pulse extraction and pulse matching have not been the focus of redundant object positioning research, and a small number of previous studies still have many problems, which have led to their failure to achieve satisfactory results in practical applications.

[0129] The present invention proposes a four-channel pulse extraction and matching algorithm for positioning redundant objects in sealed electronic equipment. The algorithm includes two stages: a two-stage variable multi-threshold pulse extraction algorithm and a threshold judgment search pulse matching algorithm, which are sequentially processed. Specifically, in the two-stage variable multi-threshold pulse extraction algorithm, it takes the average energy and average zero-crossing rate of all sampling points in the signal as reference, and sets the peak energy threshold, the endpoint energy threshold and the endpoint zero-crossing rate threshold according to the 3σ distribution criterion of the normal distribution. The pulse is pre-screened by the peak energy threshold, and the starting point and the end point of the independent pulse are accurately found by the endpoint energy threshold and the endpoint zero-crossing rate threshold set in two rounds. This effectively solves the problems existing in the existing research, such as the threshold setting rules are not fixed, the threshold reference basis is not unified, the pulse extraction reliability is not high, and continuous multiple pulses are not considered. In the threshold judgment search pulse matching algorithm, it searches for the starting time of the first pulse in the four-channel redundant object signal, and takes the starting time of each pulse in the redundant object signal captured by the acoustic emission sensor farthest from the redundant object as a reference to find a pulse group that meets the starting time threshold requirement. This effectively solves the problems of unreasonable reference object setting and unstable and unreliable pulse matching in existing research.

[0130] The four-channel pulse extraction and matching algorithm proposed in the present invention and the four combined algorithms of "three-threshold pulse extraction algorithm + sequential search pulse matching algorithm", "three-threshold pulse extraction algorithm + zero-filled time difference pulse matching algorithm", "double-threshold pulse extraction algorithm + sequential search pulse matching algorithm" and "double-threshold pulse extraction algorithm + zero-filled time difference pulse matching algorithm" were used to process 800 groups of redundant signals, and pulse groups belonging to the five algorithms were obtained respectively. On this basis, the characteristic values ​​were calculated, labels were added, and five data sets were constructed. Representative random forests and support vector machines were trained on the five data, and their classification accuracy was tested. The results show that both classifiers achieved the highest classification accuracy on the data set corresponding to the four-channel pulse extraction and matching algorithm, and the advantages were obvious. Compared with the existing research, the classification accuracy achieved by the classifier has also been greatly improved. At the same time, the two classifiers achieved outstanding and stable Precision, Recall and F1 scores on the data set corresponding to the four-channel pulse extraction and matching algorithm. Therefore, the feasibility and superiority of the four-channel pulse extraction and matching algorithm proposed in the present invention are fully verified.

[0131] The above calculation examples of the present invention are only used to explain the calculation model and calculation process of the present invention in detail, and are not intended to limit the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A multi-channel pulse extraction method for locating excess material in sealed electronic equipment, characterized in that: Pulse extraction is performed for each channel signal in a multi-pass signal used for locating redundant objects in sealed electronic equipment, comprising the following steps: Step 1: Calculate the average energy and average zero-crossing rate of the entire signal, expressed as E a and Z a , and the standard deviation of energy and the standard deviation of zero-crossing rate, denoted as σ E and σ Z ; Set the peak detection threshold to E p =E a +3σ E , set the endpoint energy threshold to E be_1 =E a -σ E , set the endpoint zero crossing rate threshold to Z be_1 =Z a -σ Z ; Step 2: Frame the signal, set the duration of each frame signal to Δt, and the starting position of the next frame signal to the end position of the previous frame signal; calculate the energy and zero-crossing rate of each frame signal; Step 3: Starting from the first frame signal, compare the energy of each frame signal with E p ; When the energy of a frame signal is greater than E p When the frame signal is recorded, the energy of each subsequent frame signal is compared with E p , until the energy of a frame signal is less than E p ; Find the frame signal with the largest energy from these frame signals, and obtain the time of the first sampling point, which is called the peak time t of the current pulse p ; Step 4: Take the peak time t of the pulse extracted in step 3 p As the starting point, the energy of each frame signal is compared with E be_1 Compare until the energy of a frame signal is found to be less than E in both directions. be_1 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the first start frame signal and the first end frame signal corresponding to the first level variable multi-threshold pulse of the current pulse; The time of the first sampling point in the first start frame signal and the first end frame signal corresponding to the first variable multi-threshold pulse of the current pulse is obtained, which is called the first candidate start time t of the current pulse. b_f1 and the first candidate end time t e_f1 ; Step 5: Take the peak time t of the pulse extracted in step 3 p As the starting point, the zero-crossing rate of each frame signal is calculated forward and backward respectively. be_1 Compare until the zero-crossing rate of a frame signal is found to be less than Z in both directions. be_1 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the second start frame signal and the second end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse; The time of the first sampling point in the second start frame signal and the second end frame signal corresponding to the first-level variable multi-threshold pulse of the current pulse is obtained, which is called the second candidate start time t of the current pulse. b_f2 and the second candidate end time t e_f2 ; Step 6: Compare the two candidate starting times t b_f1 and t b_f2 , the larger one is identified as the final starting time t b_1 ; Similarly, compare the two candidate end times t e_f1 and t e_f2 , the smaller one is considered as the final ending time t e_1 ; Step 7: Take the final end time t of the pulse extracted in step 6 e_1 The next frame signal is taken as a new starting point, and steps 3 to 6 are repeated to extract the second pulse until the last frame signal of the signal is searched; and the extracted multiple pulses are saved; Step 8: Reset the endpoint energy threshold to E be_2 =E a +σ E , set the endpoint zero crossing rate threshold to Z be_2 =Z a +σ Z ; Step 9: Based on the multiple pulses saved in step 7, the peak time t corresponding to the first pulse is p As the starting point, the energy of each frame signal is compared with E be_2 Compare until the energy of a frame signal is found to be less than E in both directions. be_2 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the first start frame signal and the first end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse; The time of the first sampling point in the first start frame signal and the first end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse is obtained, which is called the candidate start time t of the current pulse. b_s1 and the candidate end time t e_s1 ; Step 10: Take the peak time t of the first pulse saved in step 7 p As the starting point, the zero-crossing rate of each frame signal is calculated forward and backward respectively. be_2 Compare until the zero-crossing rate of a frame signal is found to be less than Z in both directions. be_2 ; Respectively identify the previous frame signals of the two frame signals as the start frame signal and the end frame signal of the current pulse, and record them as the second start frame signal and the second end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse; The time of the first sampling point in the second start frame signal and the second end frame signal corresponding to the secondary variable multi-threshold pulse of the current pulse is obtained, which is called the candidate start time t of the current pulse. b_s2 and the candidate end time t e_s2 ; Step 11: Compare the two candidate starting times t b_s1 and t b_s2 , the larger one is identified as the final starting time t b_2 ; Similarly, compare the two candidate end times t e_s1 and t e_s2 , the smaller one is considered as the final ending time t e_2 ; Step 12: starting from the second pulse saved in step 7, repeat the processing steps from step 9 to step 11 until all pulses saved in step 7 are processed; Step 13: Calculate the first duration = t e_1 -t b_1 and the second duration = t e_2 -t b_2 Calculate the ratio of the second duration to the first duration. If the ratio exceeds the duration ratio threshold, the final starting time t in step six is b_1 and the final ending time t e_1 The real starting time t of the current pulse is considered b and the actual end time t e Otherwise, the final starting time t in step 6 is b_1 and the final end time t in step 11 e_2 The real starting time t of the current pulse is considered b and the actual end time t e .

2. The multi-channel pulse extraction method for locating excess objects in sealed electronic equipment according to claim 1, characterized in that: The duration of each frame signal Δt=50 μs.

3. The multi-channel pulse extraction method for locating excess objects in sealed electronic equipment according to claim 2, characterized in that: The duration ratio threshold is 18%.

4. The multi-channel pulse extraction method for locating excess objects in sealed electronic equipment according to claim 3, characterized in that: The sampling frequency of the multi-pass signal used for locating excess objects in sealed electronic equipment is 500kHz, and the interval between sampling points is 2μs.

5. A computer storage medium, characterized in that: The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the multi-channel pulse extraction method for locating excess objects in sealed electronic equipment as described in any one of claims 1 to 4.

6. A multi-channel pulse extraction device for locating excess material in sealed electronic equipment, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the multi-channel pulse extraction method for locating excess objects in sealed electronic equipment as described in any one of claims 1 to 4.

7. A multi-channel pulse matching method for locating excess material in sealed electronic equipment, characterized in that: First, a multi-channel pulse extraction method for locating excess objects in sealed electronic equipment according to any one of claims 1 to 4 is used to extract pulses from each channel signal, and the number of multi-channels is recorded as N; then, multi-channel signal matching is performed based on the excess object signals extracted from each channel, and the matching process includes the following steps: Step 1: Obtain and save the start time of the pulse in the redundant object signal corresponding to each channel respectively; according to the start time of the first pulse in the redundant object signal corresponding to each channel, determine the redundant object signal captured by the acoustic emission sensor farthest from the redundant object, and record the number of pulses extracted by the channel corresponding to the sensor as b; Step 2: Denote the pulses contained in the excess signal determined in step 1 as P i , i = 1, 2, ..., n, the starting times of these pulses are t i ; Based on the starting time threshold T, the starting time threshold interval of each pulse is obtained [t i -T,t i ]; Step 3: Starting from P1, search for pulses with a start time within [t1-T, t1] in the redundant signals corresponding to the other N-1 channels; if pulses corresponding to N-1 channels can be found, combine them with P1 to form a pulse group; otherwise, skip P1 if the match fails; Step 4: Follow the steps in step 3 to i Pulse matching is performed to obtain multiple pulse groups, thereby completing the pulse matching of the channel redundant signal.

8. The multi-channel pulse matching method for locating excess objects in sealed electronic equipment according to claim 7, characterized in that: The starting time threshold T is 85 μs.

9. A computer storage medium, characterized in that The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the multi-channel pulse matching method for locating excess objects in sealed electronic equipment as claimed in claim 7.

10. A multi-channel pulse matching device for positioning excess material in sealed electronic equipment, characterized in that: The device comprises a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement a multi-channel pulse matching method for locating excess objects in a sealed electronic device as claimed in claim 7.

Images