A device classification alarm method based on general indicator trend analysis

Abstract

The present invention discloses a device hierarchical alarm method based on general indicator trend analysis, the steps comprising: tracing back historical data of device characteristic indicators to obtain time series trend data of the characteristic indicators; extracting data at the current moment and data within a forward set time period in the time series trend data, judging whether the data in the time period conforms to the abnormal data trend, if not, no alarm is required, if conforming, it is defined as a local trend; tracking and analyzing historical data of the current local trend, if there is no identical local trend in the historical data, performing a hierarchical alarm on the current local trend by comparing the numerical features of the current local trend with those of the historical data, if there is an identical local trend in the historical data, merging all local trends to form an overall trend, distinguishing and classifying the overall trend by numerical features and morphological features, formulating corresponding hierarchical thresholds in combination with actual alarm needs, and performing a hierarchical alarm on the overall trend.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment detection, and in particular to an equipment classification alarm method based on general indicator trend analysis. Background Art

[0002] With the development of modern industry and science and technology, as well as the improvement of industrial automation, modern mechanical production technology has evolved into large-scale, automatic, continuous, high-quality precision and various high-performance mechanical production, but some sudden failures of production process equipment are often inevitable and can easily lead to huge social and economic losses. Therefore, for continuous and fully automated enterprise production operation units, it is necessary to monitor the status of production operation equipment in real time, promptly detect signs of equipment failure, reasonably grade the failure alarm, and arrange corresponding treatment measures according to the alarm level.

[0003] The vibration and characteristic information generated by mechanical equipment during operation are the main signals reflecting the changes in mechanical equipment and its operating status. Acquiring, recording and analyzing these dynamic signals through various dynamic test instruments is the main way to monitor the status of mechanical equipment and diagnose faults. At present, the most commonly used and effective method for mechanical fault diagnosis is to use the vibration signal measured by the sensor to diagnose the mechanical status, and to extract the mechanical fault characteristic information in the vibration signal through spectrum analysis, frequency doubling analysis, sideband analysis, frequency band energy, etc.

[0004] At present, there are many studies on mechanical failure warning and alarm. For example, Si Peng et al. proposed an alarm method and system based on temperature change trend, which obtains temperature values ​​according to preset time intervals and judges alarm by comparing the temperature difference before and after (Si Peng, A temperature change trend alarm method and system, patent number: CN113970384A). Liu Tiejun et al. judged the aging degree of old batteries by comparing the time series similarity between new and old batteries (Liu Tiejun, A lithium battery aging detection method based on time series analysis, CN109613446B). Geng Bo proposed to collect historical data and conduct subsequent analysis and processing by determining whether the data has abnormal trends (Geng Bo, Abnormal trend determination method of nuclear power plant operation management process data, patent number: CN112526958A). Xue Zhiguo proposed a method for determining and monitoring alarms through monitoring thresholds. It is necessary to first determine the hardware configuration parameters of the monitoring equipment, and combine the historical operating indicator data to determine the monitoring thresholds of the equipment operating indicators, and finally issue an alarm (Xue Zhiguo, A monitoring threshold determination and monitoring alarm method, device, equipment and medium, patent number: CN113946493A).

[0005] The existing methods all have one or more of the following defects and deficiencies:

[0006] Current mechanical fault alarms rely on threshold settings or a large amount of prior knowledge. If the threshold setting requirements for the relevant thresholds are high or too narrow, frequent alarms will occur in violation of the actual operating rules of the equipment, causing on-site staff to be tired of responding; if the setting is too wide, when the threshold alarm occurs, the equipment may have been irreversibly damaged and have an irreversible impact on the factory's production. The acquisition of prior knowledge requires staff to conduct a large number of experiments or data inquiries.

[0007] At present, most intelligent mechanical signal alarms are carried out in real time, and the real-time diagnosis and analysis are carried out based on the latest signal data measured, without taking the historical signal data as the judgment condition of the alarm. Due to the great randomness and randomness of the operation of mechanical equipment, the equipment in normal operation may also measure large vibration data, and the diagnostic alarm system will give wrong conclusions in this case.

[0008] The current indicator data alarm situation not only depends on the recent numerical characteristics and morphological characteristics, but also has a great relationship with the historical indicator data characteristics. This factor is rarely taken into account in the current diagnostic alarm methods.

[0009] Existing fault diagnosis alarms are mostly simple threshold alarms, which cannot give different alarm levels according to the severity of the fault, resulting in the inability of on-site staff to arrange processing work in a timely manner according to the alarm level; nor can they give early warnings before serious problems occur in the equipment, handle the fault in advance, and prevent greater safety problems.

[0010] Some existing methods that consider historical data of equipment only refer to the amplitude information of past indicators and ignore the trend patterns of indicators in historical data, resulting in a high false alarm rate in a single dimension. Summary of the invention

[0011] In view of the problems and shortcomings of the current methods, the present invention proposes an equipment classification alarm method based on general indicator trend analysis.

[0012] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0013] A device classification alarm method based on general indicator trend analysis, comprising the following steps:

[0014] Step 1: Trace the historical data of the equipment characteristic indicators and obtain the time series trend data of the characteristic indicators;

[0015] Step 2: Extract the data at the current time in the time series trend data and the data within the forward set period, and determine whether the data within the set period conforms to the abnormal data trend. If not, it means that the device data is normal and no alarm is required. If it conforms, it means that the device data is abnormal. At this time, the abnormal data trend within the set period is defined as a local trend.

[0016] Step three: Track and analyze historical data for the current local trend. If there is no identical local trend in the historical data, perform a graded alarm on the current local trend by comparing the numerical features of the current local trend with those of the historical data. If there is an identical local trend in the historical data, combine all local trends to form an overall trend, and classify the overall trend based on the numerical and morphological features. Combined with actual alarm needs, formulate corresponding graded thresholds, and perform graded alarms on the overall trend.

[0017] A further improvement is that the characteristic indicator is a vibration signal, sound signal, temperature signal, voltage signal, current signal, pressure signal or rotation speed signal of the equipment.

[0018] A further improvement is that the types of abnormal data trends include continuous rise, continuous decline, and long-term stability.

[0019] A further improvement is that in step 2, after extracting the data at the current moment in the time series trend data and the data within the forward set time period, a data set is constructed, and the number of points in the data set that meet the abnormal data trend is determined by calculating the gradient of the data set, thereby determining whether the current data set meets the abnormal data trend.

[0020] A further improvement is that when judging whether the current data set conforms to the abnormal data trend, it is necessary to judge from the overall trend of the data and refer to the following auxiliary judgment methods:

[0021] (1) Individual points that do not meet the abnormal data trend are allowed to exist;

[0022] (2) If the rate of change of data in the data set is higher than the set rate, it is also judged to be in line with the abnormal data trend;

[0023] (3) The length of the data set can be adaptively modified according to the different morphological characteristics of different device characteristic indicators.

[0024] A further improvement is that in step 3, when the current local trend is graded and alarmed, the specific method is: calculate the proportion of the current local trend numerical characteristics exceeding the historical data numerical characteristics, and formulate corresponding graded thresholds according to the excess ratio range and the actual alarm needs, and grade the current local trend and alarm it, and refer to the following auxiliary judgment methods:

[0025] (1) For historical data, use the quartile method to eliminate abnormal data;

[0026] (2) If there are abnormal values ​​in the historical data whose gradient is greater than the set value, they will be eliminated.

[0027] A further improvement is that in step 3, the specific method of merging all local trends to form an overall trend is: identifying the local trends formed by all data sets in the historical data that meet the abnormal data trend, taking the current local trend as the starting point for identification, and gradually judging the relationship between the previous local trend and the next local trend, and merging the local trends that meet the following conditions:

[0028] (1) The numerical feature of the previous local trend must be smaller than the numerical feature of the next local trend;

[0029] (2) The time interval between the previous local trend and the next local trend remains within the set range;

[0030] When merging, the starting point of the last local trend that meets the conditions is selected as the starting point of the data set, and the end point of the current local trend is selected as the end point of the data set to construct the data set, which constitutes the overall trend.

[0031] A further improvement is that the numerical feature is amplitude, peak value, minimum value, kurtosis, skewness or margin.

[0032] A further improvement is that the morphological characteristics are fluctuations, growth, numerical value distribution or time series distribution.

[0033] The beneficial effects of the present invention are:

[0034] (1) The present invention solves the current problems of equipment fault diagnosis, which are serious reliance on indicator threshold setting, large errors in real-time monitoring data, and unclear fault alarms;

[0035] (2) The present invention proposes a method based on general indicator trend analysis, which can realize effective hierarchical alarm for equipment failure;

[0036] (3) The present invention proposes for the first time to classify trend types using local trends and overall trends in time series trend data, fully considering the historical evolution of characteristic indicators and being able to effectively reflect the severity of equipment failures;

[0037] (4) The present invention proposes for the first time to use the numerical characteristics and morphological characteristics of trend data as evaluation indicators to accurately reflect the differences between different trends;

[0038] (5) The indicator trend analysis method proposed in the present invention can effectively avoid false alarms caused by data fluctuations or measurement errors when monitoring equipment in real time;

[0039] (6) The present invention uses the characteristic relationship between current trend data and historical trend data as an evaluation index for threshold division, thereby replacing the threshold range in the conventional diagnosis system, and can overcome the dependence of the traditional diagnosis method on the characteristic index threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of the method algorithm involved in the invention;

[0041] Figure 2 This is a schematic diagram of the time series trend data of the effective value of vibration acceleration in Case 1;

[0042] Figure 3 This is a schematic diagram of the time series trend data of the effective value of the vibration velocity in Case 2;

[0043] Figure 4 This is a schematic diagram of the local trend identification results in the historical data of Case 2;

[0044] Figure 5 This is a schematic diagram of the starting and ending points of the overall trend of Case 2 in the trend time series. DETAILED DESCRIPTION

[0045] The present application is further described in detail below in conjunction with the accompanying drawings. It is necessary to point out here that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technical personnel in this field can make some non-essential improvements and adjustments to the present application based on the above application content.

[0046] like Figure 1 As shown, a device classification alarm method based on general indicator trend analysis includes the following steps:

[0047] Step 1: Trace the historical data of the equipment characteristic indicators and obtain the time series trend data of the characteristic indicators;

[0048] Step 2: Extract the data at the current moment in the time series trend data and the data within the forward set period, construct a data set, and determine the number of points in the data set that meet the abnormal data trend by calculating the gradient of the data set, so as to determine whether the current data set meets the abnormal data trend. If not, it means that the device data is normal and no alarm is required. If it meets, it means that the device data is abnormal. At this time, the abnormal data trend within the set period is defined as a local trend.

[0049] When judging whether the current data set conforms to the abnormal data trend, it is necessary to judge from the overall data trend. In addition, considering the data fluctuation of the device, it is difficult for the data set to achieve strict gradient changes. Three auxiliary judgment methods are formulated to ensure the accurate recognition rate and integrity of local trends:

[0050] (1) Individual points that do not meet the abnormal data trend are allowed to exist, called callback points, but the overall trend needs to be maintained;

[0051] (2) If the rate of change of data in the data set is higher than the set rate, it is also judged to be in line with the abnormal data trend;

[0052] (3) The length of the data set can be adaptively modified according to the different morphological characteristics of different device characteristic indicators.

[0053] Step 3: Track and analyze the historical data of the current local trend. If there is no identical local trend in the historical data, the current local trend is graded and alarmed by comparing the numerical features of the current local trend with the historical data. If there is an identical local trend in the historical data, all local trends are combined to form an overall trend. The overall trend is classified by numerical and morphological features, and the corresponding graded threshold is formulated in combination with the actual alarm needs to grade the overall trend. For example, equipment failures are divided into one, two, and three levels from minor, moderate to severe, with a total of three alarm levels.

[0054] When the current local trend is graded and alarmed, the specific method is as follows: calculate the ratio of the current local trend numerical characteristics exceeding the historical data numerical characteristics, and formulate the corresponding classification threshold according to the exceeding ratio range and the actual alarm needs, and perform graded alarm on the current local trend. Considering that there may be fluctuation values ​​in the historical data, which will affect the normal classification results, two auxiliary judgment methods are formulated for this:

[0055] (1) For historical data, use the quartile method or other statistical methods to eliminate abnormal data;

[0056] (2) If there are abnormal values ​​in the historical data whose gradient is greater than the set value, they will be eliminated.

[0057] The specific method of merging all local trends into an overall trend is to identify the local trends formed by all data sets in the historical data that meet the abnormal data trend, take the current local trend as the starting point for identification, and gradually determine the relationship between the previous local trend and the next local trend, and merge the local trends that meet the following conditions:

[0058] (1) The numerical feature of the previous local trend must be smaller than the numerical feature of the next local trend;

[0059] (2) The time interval between the previous local trend and the next local trend remains within the set range;

[0060] When merging, the starting point of the last local trend that meets the conditions is selected as the starting point of the data set, and the end point of the current local trend is selected as the end point of the data set to construct the data set, which constitutes the overall trend.

[0061] In the present invention, the type of signal processed is not limited, that is, the characteristic indicator is a vibration signal, sound signal, temperature signal, voltage signal, current signal, pressure signal or rotation speed signal of the equipment.

[0062] In the present invention, it is proposed to use local trends to reflect the change characteristics of a certain data set. The present invention does not limit the specific characteristics of the specific abnormal data trend, which can be points of continuous rise, continuous decline, long-term stability or other common change characteristics.

[0063] The present invention proposes to use the numerical characteristics and morphological characteristics of trend data as evaluation indicators. The present invention does not limit the characteristics. The numerical characteristics can be amplitude, peak value, minimum value, kurtosis, skewness or margin, etc. The morphological characteristics can be fluctuation, growth, numerical size distribution or time series distribution, etc.

[0064] In addition, the algorithm mentioned in the present invention can be implemented in the upper computer software in the form of digital signal processing, or in other ways, such as digital chips, hardware circuits, etc., to realize the calculation process.

[0065] The specific implementation method and algorithm process of this method are described in detail below in combination with two actual processing routines.

[0066] Case 1:

[0067] 1. Install the acceleration sensor on the surface of the equipment to be monitored, and sample at a constant sampling frequency to obtain the vibration signal. With a sampling interval of half an hour, obtain the effective value of the vibration acceleration data of the equipment to be monitored. By tracking for a long period of time, the time series trend data of the effective value of the vibration acceleration can be obtained. In this case, a total of 1115 acceleration effective value measurement results are selected as the time series trend data. For details, see Figure 2 .

[0068] 2. This method tentatively sets the length of the data set for the device to be 15 and the number of callback points to be 3. The current moment data and the data of the previous 14 measurement results are selected to construct the data set A0(i). The data set is differentiated to obtain the data set D0(i) and the number of callback points. Then the method in the above method steps is used to determine whether A0(i) meets the characteristic requirements of the local trend.

[0069] 3. Continue to trace the historical data of a period of time before the current data set, construct the data set B0(i), and use the above steps to calculate and identify all the local trends that meet the conditions in B0(i). After calculation, B0(i) does not contain any local trends that meet the conditions.

[0070] 4. Perform a local trend classification alarm according to the above steps, calculate the amplitude of the historical data B = 1.154, the amplitude of the local trend A = 2.654, and the ratio of the local trend amplitude to the historical data amplitude s = 1.29.

[0071] 5. Based on the above calculation results, it can be found that the vibration acceleration of the equipment remains stable in the historical data. Only the local trend A0(i) causes the vibration acceleration to rise abnormally continuously due to equipment failure, and the amplitude is large and the growth rate is high. After setting the threshold based on the actual alarm needs, the local trend is judged as a level 3 alarm.

[0072] Case 2:

[0073] 1. The test signal in this case is the effective value of vibration velocity. The effective value of vibration velocity data of the equipment to be monitored is obtained at a sampling interval of half an hour. After tracking for a period of time, the time series trend data of the effective value of vibration velocity can be obtained. In this case, a total of 745 speed effective value measurement results are selected as time series trend data. For details, see Figure 3 .

[0074] 2. Set the length of the data set to 15 and the number of callback points to 3. Select the current data and the data of the previous 14 measurements to construct the data set A1(i). Difference the data set to obtain the data set D1(i). Then use the method in the above steps to calculate and judge. A1(i) meets the characteristic requirements of the local trend.

[0075] 3. Continue to trace the historical data of the previous period of time of the A1(i) data set, construct the data set B1(i), and take the step method to calculate and identify all the local trends that meet the conditions in B1(i). After calculation, there are 7 local trends that meet the conditions in B1(i). For details, see Figure 4 The arrow mark in .

[0076] 4. According to the trend merging requirements in the method steps, only the last local trend meets the conditions and can be merged with the local trend A1(i). The starting and ending points of the merged overall trend are as follows: Figure 5 shown.

[0077] 5. Calculate the amplitude, peak-to-peak value, minimum value, fluctuation, growth, value distribution, etc. of the overall trend, and calculate Figure 5The amplitude, fluctuation, growth, value distribution, time series distribution, etc. of the historical trend data in a period of time before the starting point of the overall trend.

[0078] 6. Based on the above analysis results, it can be found that the vibration velocity index of the equipment has a large number of fluctuations in the historical data, and similar local trends have also appeared within a very short time interval from the local trend A1(i). This shows that the local trend A1(i) is not a continuous increase in the effective value of the speed caused by a sudden equipment failure, and A1(i) has no significant increase in amplitude compared to the most recent local trend. After comprehensively considering the size of other indicators, data forms and actual working conditions, the local trend A1(i) is defined as a level 1 alarm.

[0079] From the processing results of the above two cases, it can be seen that the final alarm of this method comprehensively considers the historical data of the characteristic indicators, avoiding false alarms caused by real-time measurement fluctuations. At the same time, the setting of the alarm threshold is not a single threshold range, but based on the changing trend of the characteristic indicators in the time series, the numerical features and morphological features are extracted for comprehensive graded alarm, which ultimately ensures the stability, accuracy and comprehensiveness of the graded alarm results.

[0080] In general, the main contribution of the present invention lies in the analysis of time series trend data, but the core analysis point is not to judge whether there is data anomaly by deviation from historical data, but to conduct a comprehensive analysis through the morphological characteristics and numerical characteristics of trend data, which is a point not involved in traditional diagnostic methods.

[0081] Among them, the application of morphological characteristics is reflected in the following aspects:

[0082] (1) The premise of the present invention to alarm data is to identify local trends or overall trends, and the core method of identifying these two trends is to fully consider the changes in the morphological characteristics of a section of trend data, rather than simply counting the peak value, amplitude and other characteristic parameters of historical data. Figure 4 Taking (the results of local trend identification in historical data of Case 2) as an example, the historical data contains 7 local trends. However, after fully combining the time distribution intervals and numerical size relationships of the local features, it is determined that only the last local trend meets the conditions and can be merged into the overall trend.

[0083] (2) When making an alarm determination for the overall trend, the present invention does not merely analyze the numerical characteristics of the historical data, but considers the overall trend itself as well as the fluctuation, growth, and numerical distribution of the historical data before determining the final alarm level.

[0084] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A device classification alarm method based on general indicator trend analysis, characterized in that the steps include: Step 1: Trace the historical data of the equipment characteristic indicators and obtain the time series trend data of the characteristic indicators; Step 2: Extract the data at the current time in the time series trend data and the data within the forward set period, and determine whether the data within the set period conforms to the abnormal data trend. If not, it means that the device data is normal and no alarm is required. If it conforms, it means that the device data is abnormal. At this time, the abnormal data trend within the set period is defined as a local trend. Step 3: Track and analyze the historical data of the current local trend. If there is no identical local trend in the historical data, perform a graded alarm on the current local trend by comparing the numerical features of the current local trend with those of the historical data. If there is an identical local trend in the historical data, combine all local trends to form an overall trend, and classify the overall trend based on the numerical features and morphological features. Combined with the actual alarm requirements, formulate corresponding graded thresholds, and perform graded alarm on the overall trend. In step 2, after extracting the data at the current moment and the data within the forward set period in the time series trend data, a data set is constructed, and the number of points in the data set that meet the abnormal data trend is determined by calculating the gradient of the data set, thereby determining whether the current data set meets the abnormal data trend; When judging whether the current data set conforms to the abnormal data trend, it is necessary to judge from the overall trend of the data and refer to the following auxiliary judgment methods: (1) Allow for individual points that do not meet the abnormal data trend; (2) If the rate of change of data in the data set is higher than the set rate, it is also judged to be in line with the abnormal data trend; (3) The length of the data set can be adaptively modified according to the different morphological characteristics of different device characteristic indicators; In step 3, when the current local trend is graded and alarmed, the specific method is as follows: calculate the ratio of the current local trend numerical characteristics exceeding the historical data numerical characteristics, and formulate the corresponding graded threshold according to the excess ratio range and the actual alarm requirements, and grade the current local trend and alarm it, and refer to the following auxiliary judgment methods: (1) For historical data, use the quartile method to eliminate abnormal data; (2) If there are abnormal values ​​in the historical data whose gradient is greater than the set value, they will be eliminated; In step 3, the specific method of merging all local trends to form the overall trend is: identify the local trends formed by all data sets in the historical data that meet the abnormal data trend, take the current local trend as the starting point for identification, and gradually move forward to determine the relationship between the previous local trend and the next local trend, and merge the local trends that meet the following conditions: (1) The numerical feature of the previous local trend must be smaller than the numerical feature of the next local trend; (2) The time interval between the previous local trend and the next local trend remains within the set range; When merging, the starting point of the last local trend that meets the conditions is selected as the starting point of the data set, and the end point of the current local trend is selected as the end point of the data set to construct the data set, and the data set is used to form the overall trend; By combining the time distribution interval and numerical value relationship of local features, it is determined that when only the last local trend meets the conditions, the overall trend is merged; The final alarm level is determined by considering the overall trend itself as well as the fluctuation, growth, and value distribution of historical data.

2. The device classification alarm method based on general indicator trend analysis according to claim 1 is characterized in that: The characteristic indicator is a vibration signal, a sound signal, a temperature signal, a voltage signal, a current signal, a pressure signal or a rotation speed signal of the equipment.

3. The device classification alarm method based on general indicator trend analysis according to claim 1 is characterized in that: The types of abnormal data trends include continuous increase, continuous decrease, and long-term stability.

4. The device classification alarm method based on general indicator trend analysis according to claim 1 is characterized in that: The numerical feature is amplitude, peak value, minimum value, kurtosis, skewness or margin.

5. The device classification alarm method based on general indicator trend analysis according to claim 1 is characterized in that: The morphological characteristics include fluctuation, growth, numerical value distribution or time series distribution.

Images