Working condition abnormity early warning algorithm based on difference abnormal value statistics

Abstract

The invention discloses a working condition abnormity early warning algorithm based on differential abnormal value statistics, and belongs to the field of random signal analysis. The method determines a key judgment basis of a risk by analyzing a periodical rule from operation to stop of a machine in a working state and common machine fault data; according to the method, data are acquired, related information capable of predicting abnormal working conditions is extracted from the data by utilizing an algorithm, abnormal fluctuation is screened and detected through difference results on the basis of extracting effective information, statistical analysis is carried out on the abnormal fluctuation, and based on the analysis, a method for judging risk levels in real time is established, so that graded early warning can be realized, and the working conditions can be predicted in various industrial environments. According to the method, risks of different degrees can be accurately warned in advance, the operation efficiency can be improved, the construction period delay can be reduced, and the application of the technology has important significance in improving the stability of machine operation and reducing risks and economic losses caused by manual operation errors or environmental factors.

Description

Technical Field

[0001] The present invention belongs to the technical field of construction safety prevention and control, and specifically relates to a method for analyzing working conditions and providing abnormal warning through time series difference. Background Art

[0002] In modern industrial production and operation, accurate abnormal working condition warning and fault detection are crucial to ensure the normal operation of equipment and the continuity of production. However, due to the complexity and variability of working conditions at construction sites, traditional fault detection methods based on rules or statistical methods often cannot meet the needs of practical applications.

[0003] In recent years, with the rapid development of big data and machine learning technology, data-driven anomaly detection methods have gradually attracted attention. Machine learning requires a large amount of data as support. It may achieve good results in the same field and the same construction scene, but the effect may drop sharply after changing the construction scene, resulting in missed reports and false reports. The trained data set can only be used in the same field, and retraining is required when changing the field.

[0004] This patent starts with the universality of the algorithm, and uses the idea of ​​difference to reduce the impact of numerical changes in different construction scenarios. At the same time, by setting thresholds for differential signals for screening, it achieves accurate extraction of outliers and has high robustness. By setting the alarm level, a graded warning effect with the same warning sensitivity can be achieved in different construction scenarios. Summary of the invention

[0005] The present invention proposes a method for filtering and analyzing input signals, screening out effective differential abnormal value information, and providing graded early warning for risks existing in working conditions, thereby improving the accuracy of early warning.

[0006] In an embodiment of the present invention, input signal filtering and analysis refers to filtering the input signal, such as pressure, intensity, temperature and other data, and the input signal is averaged with the smooth signal before input to obtain a smooth signal. The smooth signal is then differentially processed with the previous smooth signal to obtain a differential signal.

[0007] In an embodiment of the present invention, the larger value of the differential signal represents the huge data changes caused by the start and stop of the machine during the construction process, which needs to be filtered out; the smaller value represents the normal jitter during the operation of the machine or the error within a reasonable range of the acquisition equipment; the remaining value is the differential anomaly value, which represents the abnormal fluctuation of the machine during operation, such as abnormal changes in the collected data due to staff operating errors or damage to machine parts.

[0008] In the embodiment of the present invention, the method of graded warning for abnormal data is divided into level 1 alarm, level 2 alarm and level 3 alarm from severe to mild. In actual working conditions, level 3 alarm requires relevant personnel to check other machine data in real time and observe closely until the alarm disappears; level 2 alarm requires experts to analyze the machine data on site and determine whether to stop the equipment as soon as possible; level 1 alarm requires immediate stopping of the equipment. The alarm degree can be graded and warned by analyzing the frequency of abnormal fluctuation values.

[0009] Based on the above, the input signals described in the embodiments of the present invention are first smoothed and filtered, and then differentially processed, and then the differential signals are screened to obtain abnormal fluctuation values, and finally graded warnings are performed by analyzing the frequency of occurrence of abnormal fluctuation values. Since the differential signal can effectively avoid the warning error caused by the difference in input data under different working conditions, it has high robustness under different working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of the overall signal processing process of the present invention.

[0011] Figure 2 It is a schematic diagram of the input signal filtering process.

[0012] Figure 3 It is a schematic diagram of the process of calculating the differential value of the filtered signal.

[0013] Figure 4 It is a schematic diagram of the process of screening and warning the differential signal. DETAILED DESCRIPTION

[0014] The signal processing process will now be described in detail with reference to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings.

[0015] An abnormal working condition warning algorithm based on differential abnormal value statistics includes four processes: input signal smoothing, smoothed signal difference, differential data screening and differential data statistical warning. The overall process is shown in the attached figure. Figure 1 As shown. First, the input signal is smoothed, and then the difference between it and the previous value is calculated. The difference data needs to filter out the larger fluctuations caused by the machine stopping and the smaller fluctuations within the allowable range of the machine running. The remaining abnormal fluctuation signals are counted and graded alarms are issued.

[0016] The signal smoothing process is shown in the attached Figure 2As shown, the smoothing method is mean filtering. First, the queue is initialized with 0, and then each input Din (operating condition data) signal is queued. According to the first-in-first-out characteristics of the queue itself, the last data will be squeezed out, and then the average value of all elements in the queue is calculated by the mean, and the output is Dout (smoothed signal). During the calculation process, the length of the queue remains unchanged, and each smoothed data calculated is the average of this data and the previous input data. This step effectively filters out the points where mutations occur in the input data, making the overall signal more stable and reducing the possibility of program misjudgment.

[0017] The signal smoothing and signal differentiation process is as follows: Figure 3 As shown, for the first input data, the differential front term is initialized to 0. After the initialization is completed, the smoothed signal Dout output by the smoothing process is received. In each step, the current filtered signal is subtracted from Pre (last input) to obtain Diff (differential signal), and the differential signal is output to the next layer. After the differential calculation is completed, the value of Dout needs to be assigned to Pre as the subtrahend for the next differential calculation. Using differential signal analysis can increase the robustness of the system and avoid false alarms and omissions caused by the different maximum values ​​of the data rise in different working conditions.

[0018] The differential screening process is as follows Figure 4 As shown on the left, High and Low represent the screening thresholds, respectively. For different working conditions of the same machine, this value remains unchanged. When the machine starts running, the input signal will suddenly increase and then stabilize. In the performance of the differential signal, multiple large values ​​will suddenly appear and then decrease to normal values. Therefore, this part of the data needs to be separated from the abnormal fluctuation value; when the machine is running stably, the smaller fluctuations in the system also need to be distinguished from the abnormal fluctuations. The difference values ​​of the normal large fluctuations and the small fluctuations during operation in the differential signal are all set to 0, and the remaining values ​​are abnormal differential values.

[0019] The differential data statistical early warning process is as follows: Figure 4 As shown on the right, the system first uses a sliding window to count the Value (frequency of abnormal values), and then inputs the Value into the graded warning system. The graded system sets three thresholds from high to low crisis levels, L1 (level 1 alarm threshold), L2 (level 2 alarm threshold), and L3 (level 3 alarm threshold). When Diff exceeds any of these three thresholds, the system will sound an alarm with the highest alarm that meets the alarm conditions.

Claims

1. An abnormal working condition warning algorithm based on differential abnormal value statistics, characterized by: The steps include: Step 1: filter the input signal and use a window to perform sliding average filtering; Step 2, performing differential processing on the signal filtered in step 1; Step 3, select the differential signal in step 2 to exclude large fluctuations and small fluctuations; Step 4, perform sliding statistics on the differential signal processed in step 3, set a specific threshold and determine the alarm signal.

2. The abnormal working condition warning algorithm based on differential abnormal value statistics according to claim 1 is characterized in that: In step 1, a suitable filter window length is selected to process the data, and the data should be padded before the first signal is input.

3. The abnormal working condition warning algorithm based on differential abnormal value statistics according to claim 1 is characterized by: In the step 2, Step 2.1, data needs to be filled before the first differential signal input; Step 2.2, subtract the previous average value from the current input value to obtain the differential data; In step 2.3, the previous average value is saved for each input to calculate the differential signal.

4. The abnormal working condition warning algorithm based on differential abnormal value statistics according to claim 1 is characterized in that: In step 3, Step 3.1, filter out the points whose average value is lower than the set threshold, set their difference value to 0, and ignore the machine downtime; Step 3.2, filter out the points whose difference values ​​are greater than the set threshold, set their difference values ​​to 0, and exclude the drastic changes in data when the machine goes from stopping to running; Step 3.3, filter out the points whose difference values ​​are less than the set threshold, set their difference values ​​to 0, and exclude minor data fluctuations in normal operation.

5. The abnormal working condition warning algorithm based on differential abnormal value statistics according to claim 1 is characterized by: In step 4, Step 3.1, before the first signal is input, the sliding window needs to be filled with data; Step 3.2, set a fixed sliding window size and count the difference values ​​that are not 0 in the window; Step 3.3, set different levels of warning signals by counting and giving warning values.

Images