Automatic detection device and detection method for fan manufacturing

Abstract

The present invention relates to the technical field of fan operation data processing, and in particular to an automatic detection device and detection method for fan manufacturing, the method comprising: obtaining dimension detection data of each preset dimension collected by the fan to be detected at each preset moment in the current time period, and constructing a target data point; determining the data abnormality factor at each preset moment; screening out candidate abnormal moments, and screening out a preset number of candidate abnormal moments with the shortest time between each candidate abnormal moment; determining the possible noise interference index of the operation at the candidate abnormal moment; correcting the possible noise interference index of the operation at the candidate abnormal moment; correcting the preset sampling weight; and detecting the fan to be detected through a random forest model according to the target sampling weight. The present invention performs data processing on the dimension detection data, adaptively adjusts the sampling weight, realizes fan detection, and improves the accuracy of fan detection.

Description

Technical Field

[0001] The present invention relates to the technical field of fan operation data processing, and in particular to an automatic detection device and a detection method for fan manufacturing. Background Art

[0002] During the fan manufacturing process, it is often necessary to perform abnormality detection on the fan so that the manufactured fan can be used normally. At present, the method commonly used for abnormality detection is: using a random forest model to perform abnormality detection on the collected data. However, when using a random forest model to perform abnormality detection on the collected data related to the fan operation to achieve abnormality detection of the fan, the following technical problems often occur:

[0003] In the process of using the random forest model for anomaly detection, the collected data related to the fan operation is often sampled by random sampling, resulting in the same sampling weights for different data. However, the effective conditions of different data are often different. For example, some data may be affected by large noise. These data are often interference data and should be sampled less frequently. Therefore, in the process of using the random forest model for anomaly detection, if the same sampling weights are used to sample all data, some noise interference data may be sampled repeatedly, resulting in poor accuracy in the final fan detection. Summary of the invention

[0004] In order to solve the technical problem of poor accuracy in fan detection due to unreasonable sampling weight setting, the present invention proposes an automatic detection device and a detection method for fan manufacturing.

[0005] In a first aspect, the present invention provides an automatic detection method for fan manufacturing, the method comprising:

[0006] Obtain the dimension detection data of each preset dimension collected at each preset time of the fan to be detected in the current time period, and form the target data point with all the dimension detection data collected at the same preset time;

[0007] According to the dimension detection data and data changes under each preset dimension collected in the current time period, determine the data anomaly factor at each preset moment;

[0008] According to the data anomaly factor, candidate anomaly moments are screened out from all preset moments, and a preset number of candidate anomaly moments with the shortest duration between each candidate anomaly moment are screened out from all candidate anomaly moments to obtain a reference moment sequence corresponding to each candidate anomaly moment;

[0009] Determine the possible noise interference index of the operation at each candidate abnormal moment according to the time length between each candidate abnormal moment and each moment in the corresponding reference moment sequence, and the difference between the data abnormal factor at each candidate abnormal moment and the data abnormal factor at each moment in the corresponding reference moment sequence;

[0010] According to all dimensional detection data collected at all moments in the reference moment sequence corresponding to each candidate abnormal moment, the possible noise interference index of the operation at each candidate abnormal moment is corrected to obtain the target noise index at each candidate abnormal moment;

[0011] According to the target noise index at each candidate abnormal moment, the preset sampling weight is modified to obtain the target sampling weight corresponding to the target data point at each candidate abnormal moment;

[0012] According to the target sampling weight, the fans to be detected are detected through the random forest model.

[0013] Optionally, determining the data anomaly factor at each preset moment according to the dimension detection data and data changes under each preset dimension collected in the current time period includes:

[0014] The dimension detection data under each preset dimension collected at all preset moments in the current time period are used to form a dimension detection data sequence under each preset dimension;

[0015] Filter out dimension detection data adjacent to each dimension detection data from the dimension detection data sequence to which each dimension detection data belongs, and obtain an adjacent data set corresponding to each dimension detection data;

[0016] The collection time of each dimension detection data is used as the horizontal coordinate, and the detection data of each dimension is used as the vertical coordinate to construct the coordinate point corresponding to each dimension detection data;

[0017] The slope of the line connecting the coordinate point corresponding to each dimension detection data and the coordinate point corresponding to each dimension detection data in its corresponding adjacent data set is used as the initial slope to obtain the initial slope set corresponding to each dimension detection data;

[0018] The average of the initial slopes in the initial slope set corresponding to each dimension detection data is determined as the target slope corresponding to each dimension detection data;

[0019] The data anomaly factor at each preset moment is determined based on the dimension detection data under all preset dimensions collected at each preset moment and their corresponding target slopes.

[0020] Optionally, the formula corresponding to the data anomaly factor at a preset time is:

[0021]

[0022] in, is the data anomaly factor at the i-th preset moment; i is the sequence number of the preset moment in the current time period; norm() is the normalization function; and are the fluctuation change index and target deviation index at the i-th preset moment respectively; N is the number of preset dimensions; n is the sequence number of the preset dimension; k in is the target slope corresponding to the dimension detection data under the nth preset dimension collected at the i-th preset time; || is the absolute value function; F in is the dimension detection data of the nth preset dimension collected at the i-th preset time; F n It is the mean value of the dimension detection data collected under the nth preset dimension at all preset times in the current time period; |F in -F n | is the initial deviation factor corresponding to the dimension detection data under the nth preset dimension collected at the i-th preset time; ΔF n,max It is the maximum value among the initial deviation factors corresponding to the dimension detection data under the nth preset dimension collected at all preset times in the current time period.

[0023] Optionally, the step of selecting candidate abnormal moments from all preset moments according to the data abnormality factor includes:

[0024] If the data anomaly factor at the preset moment is greater than the preset anomaly threshold, the preset moment is determined as a candidate anomaly moment.

[0025] Optionally, determining the possible noise interference index of the operation at each candidate abnormal moment according to the time length between each candidate abnormal moment and each moment in the corresponding reference moment sequence, and the difference between the data abnormality factor at each candidate abnormal moment and the data abnormality factor at each moment in the corresponding reference moment sequence, comprises:

[0026] All candidate abnormal moments are sorted in chronological order to obtain a candidate abnormal moment sequence, and the sequence number of each candidate abnormal moment in the candidate abnormal moment sequence is determined as an initial sequence number corresponding to each candidate abnormal moment;

[0027] Any candidate abnormal moment is determined as the marking candidate moment, and the possible noise interference index of the operation at the marking candidate moment is determined according to the duration between the marking candidate moment and each candidate abnormal moment in the corresponding reference moment sequence, the difference between the initial sequence numbers corresponding to the marking candidate moment and each candidate abnormal moment in the corresponding reference moment sequence, and the difference between the data abnormality factor at the marking candidate moment and each candidate abnormality moment in the corresponding reference moment sequence.

[0028] Optionally, the formula corresponding to the possible noise interference index of the operation at the candidate time is marked as:

[0029]

[0030] Among them, δ is the possible noise interference index of the operation at the candidate time; exp() is an exponential function with a natural constant as the base; A is a preset number; δ a,a+1,1 is the relative period difference between the ath candidate abnormal moment and the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; δ a,a+1,2 and δ a,a+1,3 is δ a,a+1,1 The weight of || is the absolute value function; t is the candidate time for marking; t a and t a+1 are the ath candidate abnormal moment and the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; R is the initial sequence number corresponding to the marked candidate moment; R a and R a+1 are the initial sequence number corresponding to the ath candidate abnormal moment and the initial sequence number corresponding to the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; is the data anomaly factor at the marked candidate time; and are the data anomaly factor at the ath candidate abnormal moment and the data anomaly factor at the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; |tt a | and |tt a+1 | represents duration.

[0031] Optionally, the possible noise interference index of the operation at each candidate abnormal moment is corrected according to all dimensional detection data collected at all moments in the reference moment sequence corresponding to each candidate abnormal moment to obtain the target noise index at each candidate abnormal moment, including:

[0032] The dimension detection data under each preset dimension collected at all preset moments in the current time period are used to form a dimension detection data sequence under each preset dimension;

[0033] The absolute value of the Pearson correlation coefficient between the dimension detection data sequences under each two preset dimensions is determined as the target correlation coefficient between the two preset dimensions;

[0034] Determine any candidate abnormal moment as a marked candidate moment, and combine the marked candidate moment and its corresponding reference moment sequence to form a target moment sequence corresponding to the marked candidate moment;

[0035] The dimension detection data under each preset dimension collected at all candidate abnormal moments in the target moment sequence corresponding to the marked candidate moment constitute a sub-data sequence of the marked candidate moment under each preset dimension;

[0036] Determine the absolute value of the Pearson correlation coefficient between the sub-data sequences of the candidate marking moments in each two preset dimensions as the sub-correlation coefficient between the candidate marking moments in the two preset dimensions;

[0037] The target noise index at the marking candidate time is determined according to the sub-correlation coefficients between the preset dimensions of the marking candidate time, the target correlation coefficients between the preset dimensions and the possible noise interference index of the operation at the marking candidate time.

[0038] Optionally, the formula corresponding to the target noise index at the marking candidate time is:

[0039]

[0040] H nm1 =max((ρ nm -ρ nm1 ),0);

[0041]

[0042] Where H is the target noise index at the candidate marking time; norm() is the normalization function; δ is the possible noise interference index at the candidate marking time; N is the number of preset dimensions; n and m are the serial numbers of different preset dimensions; H nm1 H is the correlation reduction factor between the nth preset dimension and the mth preset dimension of the marking candidate moment; nm2 Yes H nm1 The weight of; max() is the maximum value function; ρ nm is the target correlation coefficient between the nth preset dimension and the mth preset dimension; ρ nm1is the sub-correlation coefficient between the nth preset dimension and the mth preset dimension at the marking candidate moment; ρ is the cumulative value of the target correlation coefficients between all different preset dimensions.

[0043] Optionally, the preset sampling weight is modified according to the target noise index at each candidate abnormal moment to obtain the target sampling weight corresponding to the target data point at each candidate abnormal moment, including:

[0044] When the target noise index at the candidate abnormal moment is greater than the preset noise threshold, the preset value is determined as the target sampling weight corresponding to the target data point at the candidate abnormal moment;

[0045] When the target noise index at the candidate abnormal moment is less than or equal to the preset noise threshold, the target sampling weight corresponding to the target data point at the candidate abnormal moment is determined according to the target noise index at the candidate abnormal moment and the preset sampling weight, wherein the preset sampling weight is positively correlated with the target sampling weight, and the target noise index is negatively correlated with the target sampling weight.

[0046] In a second aspect, the present invention provides an automatic detection device for fan manufacturing, comprising a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement the automatic detection method for fan manufacturing.

[0047] The present invention has the following beneficial effects:

[0048] An automatic detection method for fan manufacturing of the present invention processes the dimension detection data, adaptively adjusts the sampling weight, realizes fan detection, solves the technical problem of poor accuracy of fan detection due to unreasonable sampling weight setting, and improves the accuracy of fan detection. First, in the process of collecting dimension detection data, the dimension detection data collected at different times may be affected by noise to varying degrees due to the influence of factors such as the environment. In other words, there may be noise data in the dimension detection data collected at different times. Generally speaking, in the process of using the random forest model for abnormal detection, it is often necessary to sample the noise data as little as possible. Therefore, the present invention adaptively quantifies the target sampling weight corresponding to the target data point by quantifying the target noise index, so that the data affected by different noises adaptively corresponds to a sampling weight, thereby improving the accuracy of fan detection. Secondly, the present invention comprehensively considers multiple indicators related to noise interference, such as data anomaly factor, possible noise interference index of operation and target noise index, thereby improving the adaptive quantization of the target sampling weight corresponding to the target data point, thereby improving the accuracy of fan detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A flow chart of an automatic detection method for fan manufacturing according to the present invention;

[0051] Figure 2 is a flow chart of a method for determining a data anomaly factor of the present invention;

[0052] Figure 3 The present invention is a flow chart of a method for determining a target noise index. DETAILED DESCRIPTION

[0053] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0054] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0055] refer to Figure 1 , shows the process of some embodiments of an automatic detection method for fan manufacturing of the present invention. The automatic detection method for fan manufacturing includes the following steps:

[0056] Step S1, obtaining dimension detection data of each preset dimension collected at each preset moment of the fan to be detected in the current time period, and forming a target data point with all dimension detection data collected at the same preset moment.

[0057] Among them, the fan to be detected may be a fan to be detected for abnormal operation. The end time of the current time period may be the current time. The current time is also called the current time. The duration corresponding to the current time period may be a preset duration. The preset duration may be a preset duration. For example, the preset duration may be two hours. The preset time may be a preset collection time. The duration between adjacent preset moments may be the same. For example, the duration between adjacent preset moments may be 5 seconds. The preset dimension may be a dimension related to the operating state of the fan to be detected. For example, the preset dimension may be, but is not limited to: current dimension, temperature dimension, speed dimension, noise level dimension, and air volume dimension. The dimension detection data may be a numerical value under the preset dimension. For example, if the preset dimension is the current dimension, the dimension detection data under the current dimension collected by the fan to be detected at any preset time may be the current of the fan to be detected collected at the preset time. The dimension detection data may be collected by a corresponding sensor. For example, the dimension detection data under the current dimension may be collected by a current sensor.

[0058] It should be noted that in the process of collecting dimension detection data, the dimension detection data collected at different times may be affected by noise to varying degrees due to factors such as the environment. Therefore, all dimension detection data collected at the same preset time constitute the target data point, which can facilitate the subsequent analysis of whether the dimension detection data collected at the preset time is noise data. In general, sensors can be installed on fan blades or impellers. These sensors can be used to monitor parameters such as the operating status, speed, temperature, and vibration of the fan, as well as to detect the position and rotation of the fan blades or impellers. These sensors can help monitor and control the operation of the fan, and perform maintenance and maintenance when necessary to ensure the safe operation and stable performance of the fan. Common fan sensors include speed sensors, temperature sensors, and vibration sensors. Through the monitoring of these sensors, problems that may arise during the operation of the fan can be discovered and solved in a timely manner, improving the reliability and efficiency of the equipment.

[0059] As an example, this step may include the following steps:

[0060] In the first step, the method for acquiring the dimension detection data of each preset dimension collected by the fan to be detected at each preset moment in the current time period is the same. Taking the current dimension as an example, the current sensor can be used to collect the current of the fan to be detected every 5 seconds in the current time period as the dimension detection data under the current dimension. At this time, the moment of each current collection is the preset moment. For example, the start time of the current time period can be used as the first moment of current collection in the current time period. The first moment of current collection in the current time period is also the first preset moment in the current time period.

[0061] In the second step, all dimensional detection data collected at the same preset time constitute the target data points.

[0062] For example, if there are three preset dimensions, namely: current dimension, temperature dimension and speed dimension, and the current, temperature and speed collected at a preset time are the first current, the first temperature and the first speed respectively, then (first current, first temperature, first speed) can represent a target data point.

[0063] Step S2, determining the data anomaly factor at each preset moment according to the dimension detection data and data changes under each preset dimension collected in the current time period.

[0064] It should be noted that, when the data anomaly factor at the preset moment is larger, it often indicates that the dimension detection data at the preset moment is more likely to be abnormal. Since noise data and fan operation abnormality data are both abnormal data. Therefore, when the data anomaly factor at the preset moment is larger, it often indicates that the dimension detection data at the preset moment is more likely to be noise data or fan operation abnormality data.

[0065] As an example, the flowchart for determining the data anomaly factor at a preset time can be as follows: Figure 2 As shown, the following steps may be specifically included:

[0066] Step 201 , dimension detection data under each preset dimension collected at all preset moments in the current time period are used to form a dimension detection data sequence under each preset dimension.

[0067] Among them, the dimension detection data series is a time series.

[0068] For example, taking the current dimension as an example, the method for obtaining the dimension detection data sequence under the current dimension can be: sorting the currents of the fans to be detected collected in the current time period according to the collection time sequence, and the sequence obtained at this time is the dimension detection data sequence under the current dimension.

[0069] Step 202 , filter out dimension detection data adjacent to each dimension detection data from the dimension detection data sequence to which each dimension detection data belongs, and obtain an adjacent data set corresponding to each dimension detection data.

[0070] The number of dimension detection data in the adjacent data set may be 2 or 1. The number of dimension detection data in the adjacent data set corresponding to the first dimension detection data and the last dimension detection data in the dimension detection data sequence may be 1. The number of dimension detection data in the adjacent data set corresponding to the dimension detection data other than the first dimension detection data and the last dimension detection data in the dimension detection data sequence may be 2.

[0071] For example, if a dimension detection data sequence is {first dimension detection data, second dimension detection data, third dimension detection data}, the adjacent data set corresponding to the first dimension detection data may include: second dimension detection data; the adjacent data set corresponding to the second dimension detection data may include: first dimension detection data and third dimension detection data; the adjacent data set corresponding to the third dimension detection data may include: second dimension detection data.

[0072] Step 203 , using the collection time of each dimension detection data as the horizontal coordinate and each dimension detection data as the vertical coordinate, constructing the coordinate point corresponding to each dimension detection data.

[0073] Step 204 , taking the slope of the line connecting the coordinate point corresponding to each dimensional detection data and the coordinate point corresponding to each dimensional detection data in its corresponding adjacent data set as the initial slope, to obtain an initial slope set corresponding to each dimensional detection data.

[0074] The number of initial slopes in the initial slope set corresponding to the dimension detection data may be equal to the number of dimension detection data in the adjacent data set corresponding to the dimension detection data.

[0075] Step 205: determine the average of the initial slopes in the initial slope set corresponding to each dimensional detection data as the target slope corresponding to each dimensional detection data.

[0076] Step 206 , determining the data anomaly factor at each preset moment according to the dimension detection data under all preset dimensions collected at each preset moment and the corresponding target slope.

[0077] For example, the formula corresponding to the data anomaly factor at a preset time can be:

[0078]

[0079] in, is the data anomaly factor at the i-th preset moment. i is the sequence number of the preset moment in the current time period. norm() is the normalization function. and are the fluctuation change index and target deviation index at the i-th preset moment. N is the number of preset dimensions. n is the sequence number of the preset dimension. k in is the target slope corresponding to the dimension detection data under the nth preset dimension collected at the i-th preset time. || is the absolute value function. F in is the dimension detection data of the nth preset dimension collected at the i-th preset time. nIt is the mean value of the dimension detection data collected under the nth preset dimension at all preset times in the current time period. |F in -F n | is the initial deviation factor corresponding to the dimension detection data under the nth preset dimension collected at the i-th preset time. ΔF n,max It is the maximum value among the initial deviation factors corresponding to the dimension detection data under the nth preset dimension collected at all preset times in the current time period.

[0080] It should be noted that fans, as a common household appliance, mostly have a gear control function. When the gear remains unchanged, the various parameters of the fans tend to be stable during operation, and no major mutations will occur. The parameters will not exceed a certain range, and during the fan detection process, the fan gears can be detected for abnormal operation in turn. The embodiment of the present invention can detect abnormal operation of the fan when any gear is set. Therefore, when the dimension detection data collected at a preset time has a large mutation or deviation in a single or multiple dimensions, it often indicates that the dimension detection data collected at the preset time may be abnormal, and further analysis is required to determine whether it is noise data. When k in The larger the value is, the greater the fluctuation of the dimension detection data under the nth preset dimension collected at the i-th preset time is. When the value is larger, it often indicates that the change fluctuation of the dimension detection data under multiple preset dimensions collected at the i-th preset time is larger, which often indicates that the dimension detection data collected at the i-th preset time is more likely to be noise data or abnormal fan operation data. When the value is larger, it often indicates that the relative deviation between the dimension detection data under multiple preset dimensions collected at the i-th preset moment is larger, and it often indicates that the dimension detection data collected at the i-th preset moment is more likely to be noise data or abnormal fan operation data. The abnormal state factor of the fan to be detected at the i-th preset moment can be represented. The larger the value is, the more likely it is that the dimensional detection data collected at the i-th preset moment is noise data or abnormal fan operation data.

[0081] Step S3, according to the data anomaly factor, select candidate abnormal moments from all preset moments, and select a preset number of candidate abnormal moments with the shortest duration between each candidate abnormal moment from all candidate abnormal moments to obtain a reference moment sequence corresponding to each candidate abnormal moment.

[0082] The preset number may be a preset number that is smaller than the total number of candidate abnormal moments. For example, the preset number may be 10.

[0083] It should be noted that the dimension detection data collected at the candidate abnormal moment is often noise data or fan operation abnormal data. Therefore, obtaining the reference time sequence corresponding to the candidate abnormal moment can facilitate the subsequent analysis of whether the dimension detection data collected at the candidate abnormal moment is noise data.

[0084] As an example, this step may include the following steps:

[0085] In the first step, if the data anomaly factor at the preset moment is greater than the preset anomaly threshold, the preset moment is determined as a candidate anomaly moment.

[0086] The preset abnormality threshold may be a preset threshold, for example, the preset abnormality threshold may be 0.65.

[0087] In the second step, any candidate abnormal moment is determined as the marking candidate moment, and a preset number of candidate abnormal moments with the shortest time intervals to the marking candidate moment are screened out from all candidate abnormal moments, and these preset number of candidate abnormal moments constitute a reference moment sequence corresponding to the marking candidate moment.

[0088] The reference time sequence may be a time sequence, and the reference time sequence corresponding to the candidate marking time may not include the candidate marking time.

[0089] Step S4, determining the possible noise interference index of the operation at each candidate abnormal moment according to the time length between each candidate abnormal moment and each moment in the corresponding reference moment sequence, and the difference between the data abnormality factor at each candidate abnormal moment and the data abnormality factor at each moment in the corresponding reference moment sequence.

[0090] It should be noted that, based on the data anomaly factor, it is possible to preliminarily determine which dimension detection data at preset moments may be abnormal. Fan anomaly detection often requires detection of fan operation anomalies. In order to improve the accuracy of fan operation anomaly detection, noise interference often needs to be eliminated. In the process of collecting dimension detection data, the influence of factors such as the fan's own operation may often lead to the presence of noise data in the collected data. For example, when the fan rotates, it will generate mechanical vibration and a certain amount of noise. These vibrations and noise may be transmitted to nearby sensors or acquisition systems, causing fluctuations or instability in sensor readings. Some fans may also have motors or electromagnetic components, which may cause magnetic field interference. For some vibration-sensitive sensors or low-frequency sensors, these interferences may be more significant.

[0091] Since the random forest model often uses random sampling with replacement when sampling data, if unreal noise data is repeatedly sampled into multiple sub-datasets, it may greatly affect the performance of the model and reduce the detection accuracy. The real abnormal data in the data set related to fan operation is often randomly generated due to equipment problems, but the noise abnormal data caused by the interference of the fan's own operation is often related to the operating state of the fan. The fan usually has periodic behavior when running, which is related to the working principle of the motor, the state of the blades and the load of the fan. These periodic behaviors may cause the generated noise data to have similar periodic characteristics. Therefore, the stronger the periodic characteristics of the abnormal data, the more likely it is to be noise data.

[0092] Secondly, the larger the quantified possible noise interference index of the operation at the candidate abnormal moment is, the more likely it is that the dimensional detection data collected at the candidate abnormal moment is noise data.

[0093] As an example, this step may include the following steps:

[0094] In the first step, all candidate abnormal moments are sorted in chronological order to obtain a candidate abnormal moment sequence, and the sequence number of each candidate abnormal moment in the candidate abnormal moment sequence is determined as the initial sequence number corresponding to each candidate abnormal moment.

[0095] Among them, the candidate abnormal time sequence is a time series.

[0096] It should be noted that the earlier the candidate abnormal moment is, the smaller its sequence number in the candidate abnormal moment sequence is. The sequence numbers of the candidate abnormal moments in the candidate abnormal moment sequence may start from 1 and increase by 1 in sequence.

[0097] In the second step, any candidate abnormal moment is determined as the marked candidate moment, and the possible noise interference index of the operation at the marked candidate moment is determined according to the time length between the marked candidate moment and each candidate abnormal moment in the corresponding reference moment sequence, the difference between the initial sequence numbers corresponding to the marked candidate moment and each candidate abnormal moment in the corresponding reference moment sequence, and the difference between the data abnormality factor at the marked candidate moment and each candidate abnormal moment in the corresponding reference moment sequence.

[0098] For example, the formula corresponding to the possible noise interference index of the operation at the candidate marking time can be determined as follows:

[0099]

[0100] Wherein, δ is the possible noise interference index of the operation at the candidate time. exp() is an exponential function with a natural constant as the base. A is a preset number. δa,a+1,1 It is the relative period difference between the ath candidate abnormal moment and the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment. a,a+1,2 and δ a,a+1,3 is δ a,a+1,1 The weight of || is the absolute value function. t is the candidate time for marking. a and t a+1 are the ath candidate abnormal moment and the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment. R is the initial sequence number corresponding to the marked candidate moment. a and R a+1 They are respectively the initial sequence number corresponding to the a-th candidate abnormal moment and the initial sequence number corresponding to the a+1-th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment. is the data anomaly factor at the marking candidate time. and They are the data anomaly factor at the ath candidate abnormal moment and the data anomaly factor at the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment. |tt a | and |tt a+1 | represents duration.

[0101] It should be noted that It can represent the relative time difference between the ath candidate abnormal time and the marking candidate time, and can also be understood as the relative period of abnormal data corresponding to the ath candidate abnormal time and the marking candidate time. It can represent the relative time difference between the a+1th candidate abnormal moment and the marked candidate moment, and can also be understood as the relative period of abnormal data corresponding to the a+1th candidate abnormal moment and the marked candidate moment. Therefore, δ a,a+1,1 It can characterize the relative period difference between the abnormal data collected at the ath candidate abnormal moment and the a+1th candidate abnormal moment. The smaller its value is, the more similar their abnormal periods are. This often indicates that the periodicity between the abnormal data around the candidate marked moment is stronger, and the dimension detection data collected at the candidate marked moment is more likely to be noise data. a,a+1,2 When δ is larger, it often indicates that the adjacent spacing between the abnormal data collected at the adjacent candidate abnormal moments and the abnormal data collected at the marked candidate moment is smaller, which often indicates that the abnormal data collected at the adjacent candidate abnormal moments have a greater periodic impact on the abnormal data collected at the marked candidate moment. a,a+1,3When δ is larger, it often indicates that the difference in data anomaly factors between adjacent candidate anomaly moments and the marked candidate moment is smaller, which often indicates that the abnormal data collected at adjacent candidate anomaly moments has a greater periodic impact on the abnormal data collected at the marked candidate moment. Therefore, when δ is larger, it often indicates that the dimension detection data collected at the marked candidate moment is more periodic, which often indicates that the dimension detection data collected at the marked candidate moment is more likely to be noise data.

[0102] Step S5, based on all dimensional detection data collected at all moments in the reference moment sequence corresponding to each candidate abnormal moment, correct the possible noise interference index of the operation at each candidate abnormal moment to obtain the target noise index at each candidate abnormal moment.

[0103] It should be noted that, in addition to the periodic characteristics, the dimension detection data of each dimension when the fan is running often has a certain correlation. For example, when the fan speed increases, the noise level tends to increase, and when the fan current increases, the motor temperature tends to increase. Therefore, when the data has a real abnormality, the parameters with strong correlation tend to change to a certain extent, while the noise data tends to be weak in correlation. Therefore, the correlation between the data can also be used to further correct the possible noise interference index of the operation to obtain more accurate results. Secondly, the larger the target noise index at the quantified candidate abnormal moment, the more likely the dimension detection data collected at the candidate abnormal moment is to be noise data.

[0104] As an example, the target noise index at the candidate abnormal moment can be determined as follows: Figure 3 As shown, the following steps may be specifically included:

[0105] Step 301 , the dimension detection data under each preset dimension collected at all preset moments in the current time period are used to form a dimension detection data sequence under each preset dimension.

[0106] Step 302: determine the absolute value of the Pearson correlation coefficient between the dimension detection data sequences under every two preset dimensions as the target correlation coefficient between the two preset dimensions.

[0107] Step 303: determine any candidate abnormal moment as a marked candidate moment, and combine the marked candidate moment and its corresponding reference moment sequence to form a target moment sequence corresponding to the marked candidate moment.

[0108] The target time sequence may be a time series.

[0109] For example, if the candidate marking time is 09:45:06 on April 26, 2024, and the target time sequence corresponding to the candidate marking time is {09:45:00 on April 26, 2024, 09:45:13 on April 26, 2024}, then the target time sequence corresponding to the candidate marking time can be {09:45:00 on April 26, 2024, 09:45:06 on April 26, 2024, 09:45:13 on April 26, 2024}.

[0110] Step 304 , the dimension detection data under each preset dimension collected at all candidate abnormal moments in the target moment sequence corresponding to the marked candidate moment are used to form a sub-data sequence under each preset dimension of the marked candidate moment.

[0111] For example, taking the current dimension as an example, the method for obtaining the sub-data sequence of the marked candidate moment in the current dimension can be: sorting the currents of the fan to be detected collected at all candidate abnormal moments in the target moment sequence corresponding to the marked candidate moment in order of collection time. The sequence obtained at this time is the sub-data sequence of the marked candidate moment in the current dimension.

[0112] Step 305: determine the absolute value of the Pearson correlation coefficient between the sub-data sequences of the candidate marking moments in each two preset dimensions as the sub-correlation coefficient between the candidate marking moments in the two preset dimensions.

[0113] Step 306 , determining a target noise index at the candidate marking time according to the sub-correlation coefficients between the preset dimensions at the candidate marking time, the target correlation coefficients between the preset dimensions, and the possible noise interference index of the operation at the candidate marking time.

[0114] For example, the formula for determining the target noise index at the candidate marking time may be:

[0115]

[0116] H nm1 =max((ρ nm -ρ nm1 ),0);

[0117]

[0118] Where H is the target noise index at the candidate marking time. norm() is the normalization function. δ is the possible noise interference index at the candidate marking time. N is the number of preset dimensions. n and m are the serial numbers of different preset dimensions. H nm1 H is the correlation reduction factor between the nth preset dimension and the mth preset dimension of the marking candidate time.nm2 Yes H nm1 The weight of . max() is the maximum value function. ρ nm is the target correlation coefficient between the nth preset dimension and the mth preset dimension. nm1 is the sub-correlation coefficient between the nth preset dimension and the mth preset dimension at the marking candidate moment. ρ is the cumulative value of the target correlation coefficients between all different preset dimensions.

[0119] It should be noted that when δ is larger, it often indicates that the dimension detection data collected at the candidate marking time is more periodic, and it often indicates that the dimension detection data collected at the candidate marking time is more likely to be noise data. nm1 It can represent the reduction of the correlation between the nth preset dimension and the mth preset dimension of the data around the abnormal data collected at the candidate marking time relative to the overall correlation between the two dimensions. The larger the value, the weaker the correlation performance of the abnormal data collected at the candidate marking time in the two dimensions, which often indicates that the dimensional detection data collected at the candidate marking time is more likely to be noise data. When H nm2 When the value is larger, it often indicates that the sub-correlation coefficient between the nth preset dimension and the mth preset dimension of the marking candidate moment is relatively larger, and H nm2 Can be used as H nm1 Therefore, when H is larger, it often means that the dimension detection data collected at the candidate marking time is more likely to be noise data.

[0120] Step S6, according to the target noise index at each candidate abnormal moment, the preset sampling weight is corrected to obtain the target sampling weight corresponding to the target data point at each candidate abnormal moment.

[0121] The preset sampling weight may be a pre-set sampling weight. For example, the preset sampling weight may be B is the number of all target data points, which is equal to the number of preset moments in the current time period. Both the preset sampling weight and the target sampling weight are not less than 0.

[0122] It should be noted that the target sampling weight corresponding to the target data point can represent a relatively accurate sampling weight of the target data point in the subsequent random forest model anomaly detection.

[0123] As an example, this step may include the following steps:

[0124] In the first step, when the target noise index at the candidate abnormal moment is greater than the preset noise threshold, the preset value is determined as the target sampling weight corresponding to the target data point at the candidate abnormal moment.

[0125] The preset noise threshold may be a preset threshold. For example, the preset noise threshold may be 0.75. The preset value may be a preset value of 0 or approximately 0.

[0126] It should be noted that when the target noise index at the candidate abnormal moment is greater than the preset noise threshold, it can often be determined that the dimensional detection data collected at the candidate abnormal moment is noise data. At this time, the target data point at the candidate abnormal moment is often a noise data point. In order to reduce interference with subsequent fan detection, it is often necessary to set a smaller sampling weight for the target data point at the candidate abnormal moment. Therefore, the preset value can often be set to 0.

[0127] In the second step, when the target noise index at the candidate abnormal moment is less than or equal to the preset noise threshold, the target sampling weight corresponding to the target data point at the candidate abnormal moment is determined according to the target noise index at the candidate abnormal moment and the preset sampling weight.

[0128] The preset sampling weight may be positively correlated with the target sampling weight, and the target noise index may be negatively correlated with the target sampling weight.

[0129] For example, if the target noise index at the candidate abnormal moment is less than or equal to the preset noise threshold, the formula for determining the target sampling weight corresponding to the target data point at the candidate abnormal moment can be:

[0130] Where γ is the target sampling weight corresponding to the target data point at the candidate abnormal moment. B is the number of all target data points, which is equal to the number of preset moments in the current time period. is the preset sampling weight. H is the target noise index at the marking candidate time.

[0131] It should be noted that when the target noise index at the candidate abnormal moment is less than or equal to the preset noise threshold, it often means that the dimension detection data collected at the candidate abnormal moment may be affected by a certain degree of relatively less serious noise, and there may be certain valid information in it. Therefore, these dimension detection data can participate in subsequent anomaly detection, but the originally set preset sampling weights need to be reduced.

[0132] As another example, target data points other than the target data points at the candidate abnormal moment are often not noise points, so there is no need to adjust the sampling weights corresponding to these data points. Therefore, the preset sampling weights can be used as the target sampling weight corresponding to each target data point other than the target data points at the candidate abnormal moment.

[0133] Step S7: Detect the fan to be detected by using a random forest model according to the target sampling weight.

[0134] As an example, the target sampling weight corresponding to the target data point can be used as the sampling weight of the target data point in the random forest model, and the random forest model is used to perform anomaly detection. If abnormal data is detected, it is determined that the operating state of the fan to be detected is abnormal, and an abnormal reminder message is sent to remind the staff to perform further processing, such as further inspection and maintenance. Among them, the sampling weight of the target data point in the random forest model can represent the probability of the target data point being sampled in the random forest model.

[0135] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides an automatic detection device for fan manufacturing, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the above-mentioned computer program is executed by the processor, the steps of an automatic detection method for fan manufacturing are implemented.

[0136] In summary, the present invention adaptively quantifies the target sampling weight corresponding to the target data point by quantifying the target noise index, so that the data affected by different noises adaptively corresponds to a sampling weight, thereby improving the accuracy of fan detection. Secondly, the present invention comprehensively considers multiple indicators related to noise interference, such as data anomaly factors, possible operating noise interference indicators, and target noise indicators, thereby improving the adaptive quantification of the target sampling weight corresponding to the target data point, thereby improving the accuracy of fan detection.

[0137] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An automatic detection method for fan manufacturing, characterized in that: The following steps are involved: Obtain the dimension detection data of each preset dimension collected at each preset time of the fan to be detected in the current time period, and form the target data point with all the dimension detection data collected at the same preset time; According to the dimension detection data and data changes under each preset dimension collected in the current time period, determine the data anomaly factor at each preset moment; According to the data anomaly factor, candidate anomaly moments are screened out from all preset moments, and a preset number of candidate anomaly moments with the shortest duration between each candidate anomaly moment are screened out from all candidate anomaly moments to obtain a reference moment sequence corresponding to each candidate anomaly moment; Determine the possible noise interference index of the operation at each candidate abnormal moment according to the time length between each candidate abnormal moment and each moment in the corresponding reference moment sequence, and the difference between the data abnormal factor at each candidate abnormal moment and the data abnormal factor at each moment in the corresponding reference moment sequence; According to all dimensional detection data collected at all moments in the reference moment sequence corresponding to each candidate abnormal moment, the possible noise interference index of the operation at each candidate abnormal moment is corrected to obtain the target noise index at each candidate abnormal moment; According to the target noise index at each candidate abnormal moment, the preset sampling weight is modified to obtain the target sampling weight corresponding to the target data point at each candidate abnormal moment; According to the target sampling weight, the fans to be detected are detected through the random forest model.

2. The automatic detection method for fan manufacturing according to claim 1, characterized in that: The step of determining the data anomaly factor at each preset moment according to the dimension detection data and data changes under each preset dimension collected in the current time period includes: The dimension detection data under each preset dimension collected at all preset moments in the current time period are used to form a dimension detection data sequence under each preset dimension; Filter out dimension detection data adjacent to each dimension detection data from the dimension detection data sequence to which each dimension detection data belongs, and obtain an adjacent data set corresponding to each dimension detection data; The collection time of each dimension detection data is used as the horizontal coordinate, and the detection data of each dimension is used as the vertical coordinate to construct the coordinate point corresponding to each dimension detection data; The slope of the line connecting the coordinate point corresponding to each dimension detection data and the coordinate point corresponding to each dimension detection data in its corresponding adjacent data set is used as the initial slope to obtain the initial slope set corresponding to each dimension detection data; The average of the initial slopes in the initial slope set corresponding to each dimension detection data is determined as the target slope corresponding to each dimension detection data; The data anomaly factor at each preset moment is determined based on the dimension detection data under all preset dimensions collected at each preset moment and their corresponding target slopes.

3. The automatic detection method for fan manufacturing according to claim 2, characterized in that: The formula corresponding to the data anomaly factor at the preset time is: in, is the data anomaly factor at the i-th preset moment; i is the sequence number of the preset moment in the current time period; norm() is the normalization function; and are the fluctuation change index and target deviation index at the i-th preset moment respectively; N is the number of preset dimensions; n is the sequence number of the preset dimension; k in is the target slope corresponding to the dimension detection data under the nth preset dimension collected at the i-th preset time; || is the absolute value function; F in is the dimension detection data of the nth preset dimension collected at the i-th preset time; F n It is the mean value of the dimension detection data collected under the nth preset dimension at all preset times in the current time period; |F in -F n | is the initial deviation factor corresponding to the dimension detection data under the nth preset dimension collected at the i-th preset time; ΔF n,max It is the maximum value among the initial deviation factors corresponding to the dimension detection data under the nth preset dimension collected at all preset times in the current time period.

4. The automatic detection method for fan manufacturing according to claim 1, characterized in that: The method of selecting candidate abnormal moments from all preset moments according to the data abnormality factor includes: If the data anomaly factor at the preset moment is greater than the preset anomaly threshold, the preset moment is determined as a candidate anomaly moment.

5. The automatic detection method for fan manufacturing according to claim 1, characterized in that: The method determines the possible noise interference index of the operation at each candidate abnormal moment according to the time length between each candidate abnormal moment and each moment in the corresponding reference moment sequence, and the difference between the data abnormality factor at each candidate abnormal moment and the data abnormality factor at each moment in the corresponding reference moment sequence, including: All candidate abnormal moments are sorted in chronological order to obtain a candidate abnormal moment sequence, and the sequence number of each candidate abnormal moment in the candidate abnormal moment sequence is determined as an initial sequence number corresponding to each candidate abnormal moment; Any candidate abnormal moment is determined as the marking candidate moment, and the possible noise interference index of the operation at the marking candidate moment is determined according to the duration between the marking candidate moment and each candidate abnormal moment in the corresponding reference moment sequence, the difference between the initial sequence numbers corresponding to the marking candidate moment and each candidate abnormal moment in the corresponding reference moment sequence, and the difference between the data abnormality factor at the marking candidate moment and each candidate abnormality moment in the corresponding reference moment sequence.

6. The automatic detection method for fan manufacturing according to claim 5, characterized in that: The formula corresponding to the possible noise interference index of the operation at the candidate marking time is: Among them, δ is the possible noise interference index of the operation at the candidate time; exp() is an exponential function with a natural constant as the base; A is a preset number; δ a,a+1,1 is the relative period difference between the ath candidate abnormal moment and the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; δ a,a+1,2 and δ a,a+1,3 is δ a,a+1,1 The weight of || is the absolute value function; t is the candidate time for marking; t a and t a+1 are the ath candidate abnormal moment and the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; R is the initial sequence number corresponding to the marked candidate moment; R a and R a+1 are the initial sequence number corresponding to the ath candidate abnormal moment and the initial sequence number corresponding to the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; is the data anomaly factor at the marked candidate time; and are the data anomaly factor at the ath candidate abnormal moment and the data anomaly factor at the a+1th candidate abnormal moment in the reference moment sequence corresponding to the marked candidate moment; |tt a | and |tt a+1 | represents duration.

7. The automatic detection method for fan manufacturing according to claim 1, characterized in that: According to all dimensional detection data collected at all moments in the reference moment sequence corresponding to each candidate abnormal moment, the possible noise interference index of the operation at each candidate abnormal moment is corrected to obtain the target noise index at each candidate abnormal moment, including: The dimension detection data under each preset dimension collected at all preset moments in the current time period are used to form a dimension detection data sequence under each preset dimension; The absolute value of the Pearson correlation coefficient between the dimension detection data sequences under each two preset dimensions is determined as the target correlation coefficient between the two preset dimensions; Determine any candidate abnormal moment as a marked candidate moment, and combine the marked candidate moment and its corresponding reference moment sequence to form a target moment sequence corresponding to the marked candidate moment; The dimension detection data under each preset dimension collected at all candidate abnormal moments in the target moment sequence corresponding to the marked candidate moment constitute a sub-data sequence of the marked candidate moment under each preset dimension; Determine the absolute value of the Pearson correlation coefficient between the sub-data sequences of the candidate marking moments in each two preset dimensions as the sub-correlation coefficient between the candidate marking moments in the two preset dimensions; The target noise index at the marking candidate moment is determined according to the sub-correlation coefficient between the preset dimensions of the marking candidate moment, the target correlation coefficient between the preset dimensions and the possible noise interference index of the operation at the marking candidate moment.

8. The automatic detection method for fan manufacturing according to claim 7, characterized in that: The formula corresponding to the target noise index at the candidate marking time is: H nm1 =max((ρ nm -r nm1 ),0); Where H is the target noise index at the candidate marking time; norm() is the normalization function; δ is the possible noise interference index at the candidate marking time; N is the number of preset dimensions; n and m are the serial numbers of different preset dimensions; H nm1 H is the correlation reduction factor between the nth preset dimension and the mth preset dimension of the marking candidate moment; nm2 Yes H nm1 The weight of; max() is the maximum value function; ρ nm is the target correlation coefficient between the nth preset dimension and the mth preset dimension; ρ nm1 is the sub-correlation coefficient between the nth preset dimension and the mth preset dimension at the marking candidate moment; ρ is the cumulative value of the target correlation coefficients between all different preset dimensions.

9. The automatic detection method for fan manufacturing according to claim 1, characterized in that: The preset sampling weight is modified according to the target noise index at each candidate abnormal moment to obtain the target sampling weight corresponding to the target data point at each candidate abnormal moment, including: When the target noise index at the candidate abnormal moment is greater than the preset noise threshold, the preset value is determined as the target sampling weight corresponding to the target data point at the candidate abnormal moment; When the target noise index at the candidate abnormal moment is less than or equal to the preset noise threshold, the target sampling weight corresponding to the target data point at the candidate abnormal moment is determined according to the target noise index at the candidate abnormal moment and the preset sampling weight, wherein the preset sampling weight is positively correlated with the target sampling weight, and the target noise index is negatively correlated with the target sampling weight.

10. An automatic detection device for fan manufacturing, characterized in that: The invention comprises a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement an automatic detection method for fan manufacturing according to any one of claims 1 to 9.

Images