Ultra-high-speed collected data storage method of sound online monitoring system

Abstract

The invention relates to the technical field of intelligent monitoring, in particular to an ultra-high-speed collected data storage method of a sound online monitoring system. Comprising the following steps: designing a core ultra-high-speed acquisition module in a sound online monitoring system, and capturing a sound signal in real time at an extremely high sampling frequency; judging and setting an abnormal threshold value; and designing a data automatic recording mechanism in a fixed time period before and after the abnormal condition, and automatically recording the data in the fixed time period before and after the abnormal condition. According to the design, the data cache region with loss is arranged, and a dual judgment mechanism is combined, so that ultra-high-speed data acquisition, analysis and storage are realized; not only is the data processing efficiency improved, but also the monitoring system can be ensured to stably operate under a high-load condition; by optimizing a data storage strategy, the performance bottleneck of a traditional sound monitoring system in the high-speed acquisition process is effectively solved, and a more reliable and efficient solution is provided for industrial fault detection.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and in particular to a method for storing ultra-high-speed data collected by an online sound monitoring system. Background Art

[0002] Acoustic detection technology is based on a deep understanding of the mechanisms of sound generation, propagation and reception, which is used as a basis for comprehensive evaluation of industrial objects. The system usually consists of multiple sensors that can capture acoustic signals in different frequency ranges and analyze the captured data through advanced signal processing algorithms. In the voiceprint monitoring system, the acquisition frequency of the sound sensor is usually selected between 10kHz and 20kHz, mainly to capture the rich details of the voiceprint characteristics and ensure high-quality audio data. Selecting an acquisition frequency of 10kHz to 20kHz can effectively cover the frequency range that the human ear can perceive, thereby improving the accuracy and sensitivity of monitoring. In addition, according to the Nyquist theorem, the sampling frequency should be at least twice the highest frequency of the signal to be detected, ensuring that during the voiceprint recognition process, the sensor can accurately capture all important frequency components in the sound wave, thereby improving the accuracy of voiceprint matching and fault detection. By monitoring acoustic features such as sound pressure level, frequency components and time waveforms, potential fault modes such as bearing wear, gear failure or seal leakage can be identified. Compared with traditional detection methods, acoustic detection technology has the advantages of being non-invasive and real-time. It can continuously monitor during equipment operation, reduce downtime and improve production efficiency.

[0003] However, existing sound monitoring systems generally have the problem of imbalance between system computing power and acquisition rate. On the one hand, if the sampling rate is too low, it is difficult for the system to capture some high-frequency signal characteristics, which in turn affects the accuracy of fault detection. On the other hand, if the sampling rate is too high, the continuously recorded data will quickly occupy a large amount of disk space, resulting in system performance degradation, and may even cause the system to freeze or crash. At the same time, due to real-time limitations, the status monitoring system cannot do much analysis, which makes it difficult to choose which signals to analyze, which signals to store, and which signals to ignore in the sound monitoring system.

[0004] In view of this, we propose a method for saving ultra-high-speed data collected by an online sound monitoring system, which can avoid wasting system computing resources while meeting the ultra-high sampling rate. By setting up a lossy data buffer and a dual judgment mechanism, ultra-high-speed data collection, analysis and storage can be achieved. Summary of the invention

[0005] The purpose of the present invention is to provide a method for storing ultra-high-speed data collected by an online sound monitoring system to solve the problems raised in the above-mentioned background technology.

[0006] In order to solve the above technical problems, the present invention aims to provide a method for storing ultra-high-speed data collected by an online sound monitoring system, comprising the following steps:

[0007] S1. Design the core ultra-high-speed acquisition module in the sound online monitoring system: Use advanced digital signal processing technology and high-performance high-speed sound sensors and acquisition cards to build an ultra-high-speed acquisition module to capture sound signals in real time at an extremely high sampling frequency;

[0008] S2. Abnormal threshold determination setting: By analyzing historical monitoring data, we can understand the data distribution within the normal sound range and set a reasonable threshold;

[0009] S3. Design an automatic data recording mechanism within a fixed time period before and after the abnormal situation: used to automatically record the data within a fixed time period before and after the abnormal situation to improve the analysis efficiency and post-event traceability of abnormal events.

[0010] As a further improvement of the technical solution, in the ultra-high-speed acquisition module:

[0011] High-speed sound sensor selection: CRY2110, which has a measurement range of 25-130dBA and can collect frequency range of 10Hz-20kHz to meet ultra-high-speed collection requirements;

[0012] The acquisition card of choice is IDAQ-801, which has 4 channels of synchronous sampling at 256kS / s and supports IEPE power supply.

[0013] As a further improvement of the technical solution, S2 specifically includes the following steps:

[0014] S2.1. Data collection: Collect a sufficient amount of historical sound monitoring data and clean up missing values ​​in the data; the amount of data depends on the complexity of the sound changes and the required accuracy; the larger the amount of data, the more accurate the model;

[0015] S2.2. Threshold setting: Set the initial threshold based on the statistical analysis results, and support user-defined threshold settings.

[0016] As a further improvement of the technical solution, in S2.1, the missing values ​​in the cleaned data are supplemented by a hybrid improved optimization algorithm combined with a random forest algorithm; the hybrid improved optimization algorithm is a combination of the gray wolf optimization algorithm and the gold digger optimization algorithm; the specific algorithm steps are as follows:

[0017] S2.1.1. Use Bernoulli mapping to reduce the impact of random initial population on iteration in the grey wolf algorithm:

[0018]

[0019] In the formula, a is the control parameter, and its chaotic orbit state value range is (0,1); x i Describe the position of the gray wolf at the i-th iteration, x i+1 represents the position of the gray wolf at the i+1th iteration;

[0020] S2.1.2. During hunting, the act of surrounding a target with a population is defined as follows:

[0021] X(τ+1)=X P (τ)-A·|C·X P (τ)-X(τ)|

[0022] Where τ is the number of iterations; X P (τ) is the position of the prey, X(τ) is the position of the individual gray wolf, X(τ+1) is the position of the individual gray wolf in the encirclement process, and A and C are both coefficient vectors, as expressed in the following formula:

[0023]

[0024] Where r1 and r2 are random numbers between [0,1]; a is the convergence factor, which decreases linearly from 2 to 0 during the iteration process;

[0025] S2.1.3. After the prey is rounded up, the β-wolves, δ-wolves and ω-wolves will capture the prey under the leadership of the α-wolf. The mathematical model of the gray wolf's individual pursuit of the prey is described as follows:

[0026]

[0027] Where X1, X2, and X3 represent the position update methods of α wolf, β wolf, and δ wolf respectively. α , X β , X δ Respectively represent the current positions of wolf α, wolf β, and wolf δ, C α , C β , C δ They represent the coefficient vectors of wolf α, wolf β, and wolf δ respectively, X represents the current position vector of wolf ω, and X(t+1) is the position vector after individual update;

[0028] S2.1.4, the process of the gray wolf attacking its prey is the optimization process of the algorithm, which is completed through the decreasing change of the convergence factor. The update of the convergence factor determines the optimization effect;

[0029] S2.1.5. Use the above algorithm to find the optimal combination of the maximum number of features and the number of decision trees in the random forest algorithm.

[0030] As a further improvement of the technical solution, in S2.1.4, a gold panning optimization algorithm is introduced to optimize the convergence factor, and the algorithm is as follows:

[0031] S2.1.4.1. Parameters of gold mining optimization algorithm:

[0032]

[0033] In the formula, l1 is an intermediate variable, which is used to adjust the convergence factor or other related parameters in the subsequent algorithm steps to affect the optimization behavior of the algorithm; τ is the number of iterations, τ max is the maximum number of iterations;

[0034] S2.1.4.2. Improved convergence factor iteration method:

[0035]

[0036] S2.1.4.3. Introduce the gold panning optimizer into the position update formula of the α wolf:

[0037]

[0038] C α =2×r2

[0039] In the formula, r1 and r2 are random numbers between [0,1];

[0040] S2.1.4.4. Improve the calculation formula of the position update method X2 of the β wolf:

[0041] I = round (1 + r)

[0042] X2=x best +r×(x best -I×x i,j )

[0043] In the formula, r is a random number between [0,1], x best is the optimal position, x i,j is the current position; I is obtained by rounding , which is 1 or 2.

[0044] As a further improvement of the technical solution, the random forest algorithm to make up for missing values ​​includes:

[0045] Decision trees are the basic building blocks of random forest clustering. Each leaf node of a decision tree corresponds to a classification, and non-leaf nodes correspond to a division on a certain attribute. A random forest is composed of many decision trees and there is no correlation between the decision trees. Input data into the algorithm, randomly sample data to generate multiple training sets, and each decision tree in the forest will discriminate and classify it to obtain multiple classification results. The classification result with the most votes in the classification result is the final classification result.

[0046] For the data of each channel sensor of the acquisition card, take the historical data as the original data set, randomly select n data as the training set, use the random forest algorithm to establish n prediction models, and take their average value as the final prediction value;

[0047] To judge the accuracy of the predicted values ​​of the four channels of the acquisition card, there are 4n prediction models in total. For the random forest algorithm with 4n models, it is equivalent to having 4n trees. For any sample, under the average or majority voting principle, if and only if there are more than d trees that make an incorrect judgment, the error probability is:

[0048]

[0049] Among them, e rfc is the value of the error probability, i is the number of wrong judgments, that is, the number of trees with wrong judgments, ε is the probability of a tree making an incorrect judgment, and (1-ε) is the probability of a correct judgment.

[0050] As a further improvement of the technical solution, in S2.2, the initial threshold sets two thresholds, an upper and lower limit, where:

[0051] The lower limit threshold is a threshold based on historical data, that is, a certain ratio e of the root mean square value of the historical database under normal conditions is used as the upper and lower limit floating standard. When the standard is exceeded, a level 1 alarm is prompted;

[0052] The upper threshold is based on the autocorrelation threshold, that is, the autocorrelation of the data collected in this time period and the data collected in the previous time period is used as the standard. When the autocorrelation is less than a certain ratio (1-e), a first-level alarm is prompted; when both are triggered at the same time, a second-level alarm is prompted; when the first-level alarm is triggered, the data before and after the trigger time are saved. High-speed data acquisition;

[0053] The thresholds based on historical data are set as follows:

[0054]

[0055] In the formula, X rms Represents the effective value of the sound signal, and N represents the sampling data point;

[0056] Xmax =(1+e)X rms

[0057] X min =(1-e)X rms

[0058] In the formula, X max represents the upper threshold, X min Indicates the lower threshold;

[0059] The thresholds based on autocorrelation are set as follows:

[0060]

[0061] In the formula, ρ k is the autocorrelation coefficient, Cov represents the covariance, r t 、r t-m They represent the data collected in this time period and the data collected in the previous time period, σ represents the standard deviation, and the subscripts t and tm represent the data collected in this time period and the previous time period, respectively.

[0062] As a further improvement of the present technical solution, S3 specifically includes the following steps:

[0063] S3.1. Data storage before the exception: By combining queue management and conditional structures, the system stores data before and after the trigger time. The total time length T is the window. The latest high-speed collected data is input into the cache through a lossy queue entry mechanism. In this way, the data exceeding the cache capacity will automatically replace the oldest data to ensure that important information before the abnormality occurs is not lost;

[0064] S3.2, data storage after abnormality: using threshold judgment combined with conditional structure, running several cycles after the threshold is triggered to ensure data storage after abnormal conditions;

[0065] S3.3, Double judgment mechanism: Set a double judgment strategy to first determine whether the threshold condition is always triggered, to avoid high-speed data being saved under long-term abnormal working conditions, thereby increasing the hard disk load; then through the double judgment strategy, it is ensured that even if the fault exists for a long time, the system only records the storage operation once, thereby reducing the demand for storage resources and improving the efficiency and reliability of data management.

[0066] As a further improvement of the present technical solution, in S3.3, the threshold condition is always triggered by the standard that the currently collected real-time data is always outside the preset threshold range, that is, the currently collected real-time data is always greater than the preset upper threshold limit or always less than the preset lower threshold limit.

[0067] As a further improvement of the technical solution, in S3.3, the double judgment strategy is:

[0068] First level: the currently collected real-time data is outside the preset threshold range;

[0069] The second level: the duration of the currently collected real-time data outside the preset threshold range is between 1s and 3s;

[0070] When the above two conditions are triggered in sequence, the system collects and records the abnormal data once; after the abnormality is processed, the duration is reset to zero, and if the abnormality occurs again, it is recorded again, so as to maximize the utilization of the hard disk space.

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] 1. The ultra-high-speed data collection and storage method of the sound online monitoring system aims to meet the ultra-high sampling rate requirements while maximally avoiding the waste of system computing resources; this method realizes ultra-high-speed data collection, analysis and storage by setting a lossy data buffer area and combining a dual judgment mechanism; it not only improves the efficiency of data processing, but also ensures that the monitoring system can operate stably under high load conditions;

[0073] 2. The ultra-high-speed data acquisition and preservation method of the sound online monitoring system effectively solves the performance bottleneck of the traditional sound monitoring system in the high-speed acquisition process by optimizing the data storage strategy, providing a more reliable and efficient solution for industrial fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is an exemplary method flow chart of the present invention;

[0075] Figure 2 This is an exemplary schematic diagram of the random forest algorithm of the present invention;

[0076] Figure 3 This is a main interface diagram of an exemplary sound online monitoring system in the present invention;

[0077] Figure 4 This is an exemplary sub-interface diagram of the sound online monitoring system in the present invention;

[0078] Figure 5 This is an exemplary feedback interface diagram of the sound online monitoring system in the present invention;

[0079] Figure 6 This is an exemplary data storage operation principle diagram combined with queue management in the present invention;

[0080] Figure 7 It is a schematic diagram of the operation principle of the exemplary dual judgment mechanism in the present invention. DETAILED DESCRIPTION

[0081] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0082] Example 1

[0083] like Figure 1-Figure 7 As shown, this embodiment provides a method for storing ultra-high-speed data collected by a sound online monitoring system, comprising the following steps:

[0084] S1. Design the core ultra-high-speed acquisition module in the sound online monitoring system: Use advanced digital signal processing technology and high-performance high-speed sound sensors and acquisition cards to build an ultra-high-speed acquisition module to capture sound signals in real time at an extremely high sampling frequency;

[0085] In this step, in the ultra-high-speed acquisition module:

[0086] High-speed sound sensor selection: CRY2110, which has a measurement range of 25-130dBA and can collect frequency ranges of 10Hz-20kHz, meeting ultra-high-speed collection requirements;

[0087] The acquisition card of choice is IDAQ-801, which has 4 channels of synchronous sampling at 256kS / s and supports IEPE power supply.

[0088] S2. Abnormal threshold determination setting: By analyzing historical monitoring data, we can understand the data distribution within the normal sound range and set a reasonable threshold;

[0089] In the sound monitoring system, accurately setting the abnormal threshold is the key to effectively identifying potential problems. This process involves multiple considerations and technical implementations. The setting of abnormal thresholds can timely identify equipment failures or environmental abnormalities and avoid the escalation of potential risks. The specific steps include the following:

[0090] S2.1. Data collection: Collect a sufficient amount of historical sound monitoring data and clean up missing values ​​in the data; the amount of data depends on the complexity of the sound changes and the required accuracy; the larger the amount of data, the more accurate the model.

[0091] In this embodiment, the missing values ​​in the cleaned data are supplemented by using a hybrid improved optimization algorithm combined with a random forest algorithm; the hybrid improved optimization algorithm is a combination of the gray wolf optimization algorithm and the gold digger optimization algorithm; the specific algorithm steps are as follows:

[0092] S2.1.1. Use Bernoulli mapping to reduce the impact of random initial population on iteration in the grey wolf algorithm:

[0093]

[0094] In the formula, a is the control parameter, and its chaotic orbit state value range is (0,1); x i Describe the position of the gray wolf at the i-th iteration, x i+1 represents the position of the gray wolf at the i+1th iteration;

[0095] S2.1.2. During hunting, the act of surrounding a target with a population is defined as follows:

[0096] X(τ+1)=X P (τ)-A·|C·X P (τ)-X(τ)|

[0097] Where τ is the number of iterations; X P (τ) is the position of the prey, X(τ) is the position of the individual gray wolf, X(τ+1) is the position of the individual gray wolf in the encirclement process, and A and C are both coefficient vectors, as expressed in the following formula:

[0098]

[0099] Where r1 and r2 are random numbers between [0,1]; a is the convergence factor, which decreases linearly from 2 to 0 during the iteration process;

[0100] S2.1.3. After the prey is rounded up, the β-wolves, δ-wolves and ω-wolves will capture the prey under the leadership of the α-wolf. The mathematical model of the gray wolf's individual pursuit of the prey is described as follows:

[0101]

[0102]

[0103] Where X1, X2, and X3 represent the position update methods of α wolf, β wolf, and δ wolf respectively. α , X β , X δ Respectively represent the current positions of wolf α, wolf β, and wolf δ, C α , C β , C δ They represent the coefficient vectors of wolf α, wolf β, and wolf δ respectively, X represents the current position vector of wolf ω, and X(t+1) is the position vector after individual update;

[0104] S2.1.4. The process of the gray wolf attacking its prey is the optimization process of the algorithm. This process is completed through the decreasing change of the convergence factor. The update of the convergence factor determines the optimization effect. The gold panning optimization algorithm is introduced to optimize the convergence factor. The algorithm is as follows:

[0105] S2.1.4.1. Parameters of gold mining optimization algorithm:

[0106]

[0107] In the formula, l1 is an intermediate variable, which is used to adjust the convergence factor or other related parameters in the subsequent algorithm steps to affect the optimization behavior of the algorithm; τ is the number of iterations, τ max is the maximum number of iterations;

[0108] S2.1.4.2. Improved convergence factor iteration method:

[0109]

[0110] S2.1.4.3. Introduce the gold panning optimizer into the position update formula of the α wolf:

[0111]

[0112] C α =2×r2

[0113] In the formula, r1 and r2 are random numbers between [0,1];

[0114] S2.1.4.4. Improve the calculation formula of the position update method X2 of the β wolf:

[0115] I = round (1 + r)

[0116] X2=x best +r×(x best -I×x i,j )

[0117] In the formula, r is a random number between [0,1], x best is the optimal position, x i,j is the current position; I is obtained by rounding, which is 1 or 2;

[0118] S2.1.5. Use the above algorithm to find the optimal combination of the maximum number of features and the number of decision trees in the random forest algorithm. The final prediction effect of the random forest is related to these two parameters. The missing value predicted by the artificially randomly set parameters is not as good as that obtained by the intelligent algorithm. The hybrid optimization algorithm is worse than the original algorithm.

[0119] Furthermore, the random forest algorithm can make up for missing values ​​by:

[0120] Decision tree is the basic component of random forest clustering and is an intuitive model; decision tree is a prediction model that represents a mapping relationship between object attributes and object values, such as Figure 2 As shown in the figure; because the decision tree is a tree structure, each leaf node corresponds to a classification, and the non-leaf node corresponds to a division on a certain attribute; the random forest is composed of many decision trees and there is no connection between the decision trees; input the data into the algorithm, randomly sample the data to generate multiple training sets, and each decision tree in the forest will discriminate and classify it to obtain multiple classification results. The classification result with the most votes in the classification result is the final classification result;

[0121] Assume that for the data of each channel sensor of the acquisition card, take the historical data as the original data set, randomly select 10 data of size 10,000 as the training set, use the random forest algorithm to establish 10 prediction models, and take their average value as the final prediction value;

[0122] To judge the accuracy of the predicted values ​​of the four channels of the acquisition card, there are 40 prediction models in total. For the random forest algorithm with 40 models, it is equivalent to having 40 trees. For any sample, under the average or majority voting principle, if and only if more than 20 trees make wrong judgments, the error probability is:

[0123]

[0124] Among them, e rfc is the value of the error probability, i is the number of wrong judgments, that is, the number of trees with wrong judgments, ε is the probability of a tree making an incorrect judgment, and (1-ε) is the probability of a correct judgment.

[0125] S2.2, Threshold setting: Set the initial threshold based on the statistical analysis results, and support user-defined threshold settings. The upper and lower limits can be set as multiples of the RMS value. It is necessary to select the appropriate multiple according to the actual situation, which usually requires some experiments and adjustments.

[0126] In this step, the initial threshold sets two thresholds, upper and lower limits, where:

[0127] The lower limit threshold is a threshold based on historical data, that is, a certain ratio e (preferably 20% in this solution) of the root mean square value of the historical database under normal conditions is used as the upper and lower limit floating standard. When the standard is exceeded, a first-level alarm is prompted;

[0128] The upper threshold is based on the autocorrelation threshold, that is, the autocorrelation of the data collected in this time period and the data collected in the previous time period is used as the standard. When the autocorrelation is less than a certain ratio (1-e), that is, 80%, a first-level alarm is prompted; when both are triggered at the same time, a second-level alarm is prompted; when the first-level alarm is triggered, the data before and after the trigger time are saved. High-speed data acquisition, secondary alarm triggering, while saving data, pop-up prompt window on the monitoring main interface, and analyze frequency components;

[0129] The thresholds based on historical data are set as follows:

[0130]

[0131] Where, X rms Represents the effective value of the sound signal, and N represents the sampling data point;

[0132] X max =1.2X rms

[0133] X min =0.8X rms

[0134] Where, X max represents the upper threshold, X min Indicates the lower threshold;

[0135] The thresholds based on autocorrelation are set as follows:

[0136]

[0137] In the formula, ρ k is the autocorrelation coefficient, Cov represents the covariance, r t 、r t-m They represent the data collected in this time period and the data collected in the previous time period respectively, σ represents the standard deviation, and the subscripts t and tm represent the data collected in this time period and the previous time period.

[0138] Furthermore, users are allowed to customize threshold settings according to their specific application scenarios and needs, which can improve customer satisfaction and system practicality. Custom threshold settings are as follows:

[0139] S2.2.1. Provide a user interface: Design a user-friendly interface that allows users to easily view and adjust the current threshold settings; use LabVIEW to draw the sound monitoring and acquisition interface. The main interface is as follows: Figure 3 As shown;

[0140] S2.2.2. Provide real-time adjustment methods: In the sub-interface, users can adjust the threshold in a variety of ways, such as directly entering a value, clicking a button to adjust, and adjusting the sound decibel threshold according to the construction schedule to avoid abnormal data recording caused by human interference in construction. The sub-interface on the upper part of the motor is as follows, Figure 4 As shown in the figure, the left side of the figure shows the basic information of the display device, such as sampling frequency, sampling time, channel information, etc., the middle part is the real-time waveform and spectrum of the sound signal, and the right side is the monitoring index and alarm indicator light;

[0141] S2.2.3. Provide feedback mechanism: Provide feedback mechanism to let users understand the impact of adjusting thresholds on system performance. This feedback function is implemented through conditional judgment structure combined with threshold indicator light analysis. For example, the impact of adjusted thresholds on alarm frequency can be displayed. Figure 5 As shown, when the threshold is exceeded, an alarm is first displayed on the main interface, the corresponding component indicator light turns red, and a reminder window pops up. Then the fault time is written in the report, based on which the impact of the current set threshold on the alarm frequency can be calculated.

[0142] S3. Design an automatic data recording mechanism within a fixed time period before and after the abnormal situation: used to automatically record the data within a fixed time period before and after the abnormal situation to improve the analysis efficiency and post-event tracing capability of abnormal events; specifically, the following steps are included:

[0143] S3.1. Data storage before the exception: By combining queue management and conditional structures, the real-time storage of key data is ensured; specifically, the system stores data before and after the trigger time. The total duration T is the window, and T is preferably 30 seconds. The latest high-speed acquisition data is input into the cache through a lossy queue entry mechanism. In this way, data exceeding the cache capacity will automatically replace the oldest data to ensure that important information before an abnormality occurs is not lost. Its operating mechanism is as follows: Figure 6 As shown;

[0144] S3.2, data storage after abnormality: using threshold judgment combined with conditional structure, running several cycles after the threshold is triggered to ensure data storage after abnormal conditions;

[0145] The system sets one cycle, and a cycle refers to completing a task of collecting, processing, and saving data. If one cycle takes 1 second, then in this embodiment, it can be understood as a total of 30 cycles of 15 seconds before and after, that is, 30 cycles are completed after the abnormal data triggers the threshold;

[0146] S3.3, Double judgment mechanism: Set a double judgment strategy to first determine whether the threshold condition is always triggered, to avoid high-speed data being stored under abnormal working conditions for a long time, thereby increasing the hard disk load; then through the double judgment strategy, it is ensured that even if the fault exists for a long time, the system only records the storage operation once, thereby reducing the demand for storage resources and improving the efficiency and reliability of data management. Its operating mechanism is as follows Figure 7 shown.

[0147] The threshold condition is always triggered when: the currently collected real-time data is always outside the preset threshold range, that is, the currently collected real-time data is always greater than the preset upper threshold limit or always less than the preset lower threshold limit;

[0148] In the prior art, in a normal system that relies on threshold judgment and storage, when the real-time data is always outside the standard range, the data will continue to be recorded, which will also cause the problem of rapid consumption of hard disk space.

[0149] Furthermore, the double judgment strategy is:

[0150] First level: the currently collected real-time data is outside the preset threshold range;

[0151] The second level: the duration of the currently collected real-time data outside the preset threshold range is between 1s and 3s;

[0152] When the above two conditions are triggered in sequence, the system collects and records abnormal data once;

[0153] The advantages of the dual judgment mechanism are: it can avoid sudden changes in sound caused by external artificial stimulation; at the same time, even if the data is always outside the preset threshold range, this type of abnormality is only recorded once. After the abnormality is processed, the duration is reset to zero, and it will be recorded again if it occurs again, which can maximize the utilization of hard disk space.

[0154] In addition, it is worth mentioning that, generally speaking, the acquisition frequency of sound signals is between 10k and 20k. If all data are saved, the hard disk load is too large. Therefore, only 30 seconds of data under abnormal conditions are retained for specific analysis of the abnormality afterwards. At the same time, in order to avoid long-term abnormality causing data to be kept and increase the hard disk load, an abnormal condition is set to trigger only 30 cycles. This program is designed to efficiently monitor and record the data of the unit before and after the abnormal state occurs. By combining queue management and conditional structure, the real-time preservation of key data is ensured. Specifically, the system uses 30 seconds as a window and inputs the latest high-speed acquisition data into the cache through a lossy queue entry mechanism. In this way, data that exceeds the cache capacity will automatically replace the oldest data to ensure that important information before the abnormality occurs is not lost. In terms of abnormal state monitoring, the system sets a threshold. When the monitored measurement value exceeds this threshold, the corresponding timing reset light is triggered to indicate the activation of the abnormal condition. At this time, the timing function starts to run. If the timing time remains within the range of 1-3 seconds, the system will light up the indicator light and save the current measurement data to achieve timely recording of abnormal conditions. At the same time, the cyclic trigger mechanism of data preservation ensures that key data will be recorded in the event of continuous faults or abnormal conditions to avoid information loss. The purpose of setting the above-mentioned timing reset light is to optimize the storage management of the system. Under high sampling frequency, long-term fault status monitoring may lead to rapid exhaustion of data storage. This design uses a dual judgment mechanism to ensure that even if the fault exists for a long time, the system only records the storage operation once, thereby reducing the demand for storage resources and improving the efficiency and reliability of data management. The overall system architecture effectively improves the safety monitoring capability of the unit and provides important data support for subsequent fault diagnosis and maintenance.

[0155] Those skilled in the art will appreciate that the process of implementing all or part of the steps of the above embodiments may be completed by hardware, or may be completed by instructing related hardware through a program.

[0156] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for storing ultra-high-speed data collected by an online sound monitoring system, characterized in that: The steps include: S1. Design the core ultra-high-speed acquisition module in the sound online monitoring system: Use advanced digital signal processing technology and high-performance high-speed sound sensors and acquisition cards to build an ultra-high-speed acquisition module to capture sound signals in real time at an extremely high sampling frequency; S2. Abnormal threshold determination setting: By analyzing historical monitoring data, we can understand the data distribution within the normal sound range and set a reasonable threshold; S3. Design an automatic data recording mechanism within a fixed time period before and after the abnormal situation: used to automatically record the data within a fixed time period before and after the abnormal situation to improve the analysis efficiency and post-event traceability of abnormal events.

2. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 1, characterized in that: In the ultra-high-speed acquisition module: High-speed sound sensor selection: CRY2110, with a measurement range of 25 to 130 dBA and a frequency range of 10 Hz to 20 kHz; The acquisition card of choice is IDAQ-801, which has 4 channels of synchronous sampling at 256kS / s and supports IEPE power supply.

3. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 2, characterized in that: The S2 specifically includes the following steps: S2.

1. Data collection: Collect sufficient historical sound monitoring data and clean up missing values ​​in the data; S2.

2. Threshold setting: Set the initial threshold based on the statistical analysis results, and support user-defined threshold settings.

4. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 3 is characterized in that: In S2.1, the missing values ​​in the cleaned data are supplemented by a hybrid improved optimization algorithm combined with a random forest algorithm; the hybrid improved optimization algorithm is a combination of the gray wolf optimization algorithm and the gold digger optimization algorithm; the specific algorithm steps are as follows: S2.1.

1. Use Bernoulli mapping to reduce the impact of random initial population on iteration in the grey wolf algorithm: In the formula, a is the control parameter, and its chaotic orbit state value range is (0,1); x i Describe the position of the gray wolf at the i-th iteration, x i+1 represents the position of the gray wolf at the i+1th iteration; S2.1.

2. During hunting, the act of surrounding a target with a population is defined as follows: X(τ+1) ＝ X P (t)-A·|C·X P (τ)-X(τ)| In the formula, τ is the number of iterations; X P (τ) is the position of the prey, X(τ) is the position of the individual gray wolf, X(τ+1) is the position of the individual gray wolf in the encirclement process, and A and C are both coefficient vectors, as expressed in the following formula: Where r1 and r2 are random numbers between [0,1]; a is the convergence factor, which decreases linearly from 2 to 0 during the iteration process; S2.1.

3. After the prey is rounded up, the β-wolves, δ-wolves and ω-wolves will capture the prey under the leadership of the α-wolf. The mathematical model of the gray wolf's individual pursuit of the prey is described as follows: Where X1, X2, and X3 represent the position update methods of α wolf, β wolf, and δ wolf respectively. α , X β , X δ Respectively represent the current positions of wolf α, wolf β, and wolf δ, C α , C β , C δ They represent the coefficient vectors of wolf α, wolf β, and wolf δ respectively, X represents the current position vector of wolf ω, and X(t+1) is the position vector after individual update; S2.1.4, the process of the gray wolf attacking its prey is the optimization process of the algorithm, which is completed through the decreasing change of the convergence factor; S2.1.

5. Use the above algorithm to find the optimal combination of the maximum number of features and the number of decision trees in the random forest algorithm.

5. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 4, characterized in that: In S2.1.4, the gold panning optimization algorithm is introduced to optimize the convergence factor. The algorithm is as follows: S2.1.4.

1. Parameters of gold mining optimization algorithm: In the formula, l1 is an intermediate variable; τ is the number of iterations, τ max is the maximum number of iterations; S2.1.4.

2. Improved convergence factor iteration method: S2.1.4.

3. Introduce the gold panning optimizer into the position update formula of the α wolf: C α ＝2×r2 In the formula, r1 and r2 are random numbers between [0,1]; S2.1.4.

4. Improve the calculation formula of the position update method X2 of the β wolf: I = round (1 + r) X2=x best +r×(x best -I×x i,j ) In the formula, r is a random number between [0,1], x best is the optimal position, x i,j is the current position; I is obtained by rounding , which is 1 or 2.

6. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 5, characterized in that: The random forest algorithm to make up for missing values ​​includes: Decision trees are the basic building blocks of random forest clustering. Each leaf node of a decision tree corresponds to a classification, and non-leaf nodes correspond to a division on a certain attribute. A random forest is composed of many decision trees and there is no correlation between the decision trees. Input data into the algorithm, randomly sample data to generate multiple training sets, and each decision tree in the forest will discriminate and classify it to obtain multiple classification results. The classification result with the most votes in the classification result is the final classification result. For the data of each channel sensor of the acquisition card, take the historical data as the original data set, randomly select n data as the training set, use the random forest algorithm to establish n prediction models, and take their average value as the final prediction value; To judge the accuracy of the predicted values ​​of the four channels of the acquisition card, there are 4n prediction models in total. For the random forest algorithm with 4n models, it is equivalent to having 4n trees. For any sample, under the average or majority voting principle, if and only if there are more than d trees that make an incorrect judgment, the error probability is: Among them, e rfc is the value of the error probability, i is the number of wrong judgments, that is, the number of trees with wrong judgments, ε is the probability of a tree making an incorrect judgment, and (1-ε) is the probability of a correct judgment.

7. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 6, characterized in that: In S2.2, the initial threshold sets two thresholds, upper and lower limits, where: The lower limit threshold is a threshold based on historical data, that is, a certain ratio e of the root mean square value of the historical database under normal conditions is used as the upper and lower limit floating standard. When the standard is exceeded, a level 1 alarm is prompted; The upper threshold is based on the autocorrelation threshold, that is, the autocorrelation of the data collected in this time period and the data collected in the previous time period is used as the standard. When the autocorrelation is less than a certain ratio (1-e), a first-level alarm is prompted; when both are triggered at the same time, a second-level alarm is prompted; when the first-level alarm is triggered, the data before and after the trigger time are saved. High-speed data acquisition; The thresholds based on historical data are set as follows: Where, X rms Represents the effective value of the sound signal, and N represents the sampling data point; X max =(1+e)X rms X min =(1-e)X rms In the formula, X max represents the upper threshold, X min Indicates the lower threshold; The thresholds based on autocorrelation are set as follows: In the formula, ρ k is the autocorrelation coefficient, Cov represents the covariance, r t 、r t-m They represent the data collected in this time period and the data collected in the previous time period, σ represents the standard deviation, and the subscripts t and tm represent the data collected in this time period and the previous time period, respectively.

8. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 7, characterized in that: The S3 specifically includes the following steps: S3.

1. Data storage before the exception: By combining queue management and conditional structures, the system stores data before and after the trigger time. The total duration T is the window, and the latest high-speed acquisition data is input into the buffer through a lossy queue entry mechanism; S3.2, data storage after anomaly: using threshold judgment combined with conditional structure, running several cycles after the threshold is triggered; S3.3, Double judgment mechanism: Set a double judgment strategy to first determine whether the threshold condition is always triggered, and then use the double judgment strategy to ensure that the system only records the storage operation once even if the fault exists for a long time.

9. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 8, characterized in that: In S3.3, the threshold condition is always triggered if the real-time data currently collected is always outside the preset threshold range, that is, the real-time data currently collected is always greater than the preset upper threshold limit or always less than the preset lower threshold limit.

10. The method for storing ultra-high-speed data collected by an online sound monitoring system according to claim 9, characterized in that: In S3.3, the double judgment strategy is: First level: the currently collected real-time data is outside the preset threshold range; The second level: the duration of the currently collected real-time data outside the preset threshold range is between 1s and 3s; When the above two conditions are triggered in sequence, the system collects and records the abnormal data once; after the abnormality is processed, the duration is reset to zero, and if the abnormality occurs again, it will be recorded again.

Images