Frying production line fault monitoring method based on data analysis
By screening and purifying the temperature data of the frying production line and using PCA to obtain control limits, the problems of accuracy and sensitivity in fault monitoring of the frying production line were solved, and early warning and accurate monitoring of potential faults were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-03-27
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies for fault monitoring in frying production lines suffer from insufficient accuracy and sensitivity, particularly due to misjudgments and decreased monitoring performance caused by fixed alarm thresholds and PCA models relying on historical data quality.
By analyzing historical temperature data at each monitoring point of the frying production line, data from potential healthy cycles are screened out. Control limits for the T² and SPE statistics are obtained using PCA. Combined with the synergistic relationship within the time window, the data is purified to improve monitoring accuracy and sensitivity.
It enables precise detection of faults in the frying production line, improves the accuracy and sensitivity of fault monitoring, and reduces economic losses from product scrap and equipment damage.
Smart Images
Figure CN121743958A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, and in particular to a deep learning based fault monitoring method for a frying production line. BACKGROUND
[0002] Stable and efficient operation of the frying production line is the core link in the field of food processing. The technical difficulty lies in the strong nonlinearity and anisotropy of the thermodynamic process in the frying pot, such as uneven heating pipe layout, oil flow circulation dead angle, food load dynamic change, etc., which together cause a large temperature difference in the pot area. Abnormal local oil temperature, as a typical precursor to failure, is usually manifested as a slow drift of the temperature of a specific sensor monitoring point or an abnormal increase in the temperature difference with the neighborhood. This fault is formed insidiously, and if an efficient and intelligent monitoring method is not used, when it develops into a dominant fault such as severe coking of the heating pipe and deterioration of the oil quality, it will cause serious economic losses such as product batch rejection, energy consumption surge and equipment damage.
[0003] Industry usually sets temperature sensors at key points of the frying production line and configures fixed alarm thresholds to monitor the frying pot for faults, or introduces a principal component analysis (PCA) trained model to mine the correlation between variables. However, due to the inherent high complexity of the frying production line, the requirements of different key points on the monitoring algorithm differ potentially, the fixed alarm threshold method cannot capture local temperature difference anomalies and gradual faults, and when PCA technology is directly applied, its monitoring performance depends entirely on the quality of the historical training data. If the historical data used for modeling itself contains unobserved, local early fault data, the trained model will mistakenly consider this "sick state" as "normal state", resulting in inaccurate control limits, poor monitoring sensitivity, and ultimately failing to achieve early warning.
[0004] Therefore, how to improve the accuracy and sensitivity of fault monitoring of the frying production line becomes a problem to be solved. SUMMARY
[0005] Therefore, the embodiments of the present application provide a deep learning based fault monitoring method for a frying production line to solve the problem of how to improve the accuracy and sensitivity of fault monitoring of the frying production line.
[0006] The deep learning based fault monitoring method for a frying production line provided in the embodiments of the present application comprises the following steps: Obtain oil temperature data of each monitoring point on the frying production line at each time in a preset number of historical periods to obtain a temperature data sequence of each monitoring point in each historical period, wherein the equipment on the frying production line in the historical period has no alarm record, and the fried products are all qualified. Based on the data fluctuation of the temperature data sequence of each monitoring point in each historical period, the possible health monitoring points in each historical period are obtained, and the historical period in which all monitoring points are possible health monitoring points is recorded as the possible health period. For any possible healthy cycle, based on the difference between the oil temperature data of all monitoring points in each time window of the possible healthy cycle and the oil temperature data of all monitoring points in all historical cycles, the comprehensive deviation index of the possible healthy cycle is obtained. Based on the comprehensive deviation index of each possible healthy cycle, all temperature data sequences are purified to obtain the final health data. Using PCA, control limits for the T² and SPE statistics of all final health data are obtained. Based on the real-time oil temperature data of each monitoring point, real-time T² and SPE statistics are obtained. Based on the real-time T² and SPE statistics, as well as the control limits for the T² and SPE statistics, real-time fault monitoring of the frying production line is performed.
[0007] Preferably, obtaining the possible health monitoring points within each historical period based on the data fluctuation of the temperature data sequence of each monitoring point within each historical period includes: The historical period is divided into a heating period and a heat preservation period. For any historical period, the first degree of anomaly of any monitoring point in any historical period is obtained based on the heating rate of any monitoring point in the heating period and the fluctuation of oil temperature data in the heat preservation period. Based on the overall fluctuation of the temperature data sequence of any monitoring point within any historical period, the second degree of anomaly at any monitoring point within any historical period is obtained; The weighted sum of the first degree of abnormality and the second degree of abnormality is used to obtain the weighted abnormality index of any monitoring point in any historical period. Obtain the weighted anomaly index of each monitoring point within any historical period, and select the maximum value among the weighted anomaly indices of all monitoring points within any historical period, which is recorded as the maximum weighted anomaly index of any historical period. Obtain the maximum weighted anomaly index for each historical period, and use the first preset quantile of all maximum weighted anomaly indices as the anomaly index threshold. If the weighted anomaly index of any monitoring point within any historical period is less than the anomaly index threshold, then the monitoring point is recorded as a possible health monitoring point within the historical period.
[0008] Preferably, the first abnormality degree of the any monitoring point in the any historical period is obtained according to the heating rate of the any monitoring point in the heating period and the fluctuation of the oil temperature data in the holding period in the any historical period, comprising: The temperature data sequence of the any monitoring point in the any historical period is linearly fitted to obtain a fitting straight line, and the absolute value of the slope of the fitting straight line is linearly normalized to obtain an oil temperature fluctuation trend value of the any monitoring point in the any historical period. The heating rate deviation degree is linearly normalized to obtain a heating rate deviation degree normalized value. The heating rate deviation degree normalized value and the variation coefficient normalized value are weighted and summed to obtain the first abnormality degree of the any monitoring point in the any historical period. The heating rate deviation degree normalized value and the variation coefficient normalized value are weighted and summed to obtain the first abnormality degree of the any monitoring point in the any historical period.
[0009] Preferably, the second abnormality degree of the any monitoring point in the any historical period is obtained according to the overall fluctuation of the temperature data sequence of the any monitoring point in the any historical period, comprising: The temperature data sequence of the any monitoring point in the any historical period is linearly fitted to obtain a fitting straight line, and the absolute value of the slope of the fitting straight line is linearly normalized to obtain an oil temperature fluctuation trend value of the any monitoring point in the any historical period. The heating rate deviation degree is linearly normalized to obtain a heating rate deviation degree normalized value. The heating rate deviation degree normalized value and the variation coefficient normalized value are weighted and summed to obtain the first abnormality degree of the any monitoring point in the any historical period.
[0010] Preferably, the comprehensive deviation index of the any possible health period is obtained according to the difference between the oil temperature data of all monitoring points in each time window of the any possible health period and the oil temperature data of all monitoring points in all historical periods, comprising: dividing the any possible health cycle into at least two time windows, for any time window, constructing a covariance matrix according to the oil temperature data of each monitoring point in the any time window, denoted as a local time covariance matrix; constructing a covariance matrix according to the temperature data sequences of all monitoring points in all historical cycles, denoted as a reference covariance matrix; calculating the Frobenius norm between the local time covariance matrix and the reference covariance matrix, obtaining the degree of cooperative abnormality of all monitoring points in the any time window in the any possible health cycle; obtaining the degree of cooperative abnormality of all monitoring points in each time window in the any possible health cycle, obtaining the maximum value in all degrees of cooperative abnormality, denoted as a comprehensive deviation index of the any possible health cycle.
[0011] Preferably, the comprehensive deviation index of each possible health cycle is used to purify all temperature data sequences to obtain final health data, including: a second preset quantile of the comprehensive deviation indexes of all possible health cycles is obtained, denoted as a comprehensive deviation index threshold, possible health cycles with a comprehensive deviation index less than the comprehensive deviation index threshold are recorded as final health cycles, and data in the temperature data sequence of each monitoring point in each final health cycle is recorded as final health data.
[0012] Preferably, the real-time T2 statistics and the real-time SPE statistics, and the control limit of the T2 statistics and the SPE statistics are used to perform real-time fault monitoring on the frying production line, including: if the real-time T2 statistics does not exceed the control limit of the T2 statistics, and the real-time SPE statistics does not exceed the control limit of the SPE statistics, it is determined that the frying production line does not have a fault; if the real-time T2 statistics exceeds the control limit of the T2 statistics, or the real-time SPE statistics exceeds the control limit of the SPE statistics, real-time T2 contribution maps and real-time SPE contribution maps are generated according to the real-time oil temperature data of each monitoring point, and the real-time T2 contribution maps and the real-time SPE contribution maps are used to locate the fault monitoring point in the frying production line in real time.
[0013] Compared with the prior art, the embodiment of the present application has the following beneficial effects: The application realizes the preliminary screening of the temperature data sequence of all monitoring points in all historical periods by analyzing the temperature data sequence of each monitoring point on the frying production line in each historical period, obtaining the possible health monitoring points in each historical period, and recording the historical period in which all monitoring points are possible health monitoring points as a possible health period, so as to exclude obviously unhealthy data; the synergistic relationship between all monitoring points in each time window is analyzed according to the difference between the oil temperature data of all monitoring points in each time window of the possible health period and the oil temperature data of all monitoring points in all historical periods, so as to further exclude unhealthy data in which the synergistic relationship is broken, obtain the final health data, complete the purification operation of the temperature data sequence of all monitoring points in all historical periods, and then use PCA to obtain the control limit of the T2 statistics and the SPE statistics of all final health data, so as to realize real-time fault monitoring of the frying production line, and improve the accuracy and sensitivity of fault monitoring of the frying production line. BRIEF DESCRIPTION OF DRAWINGS
[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.
[0015] Figure 1 It is a flow chart of a frying production line fault monitoring method based on data analysis provided by the first embodiment of the present application. DETAILED DESCRIPTION
[0016] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the drawings. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.
[0017] It should be noted that the terms "first", "second" and the like in the specification of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application.
[0018] In order to illustrate the technical solutions of the present application, the following will be described by specific embodiments.
[0019] Reference Figure 1, is a method flow chart of a frying production line fault monitoring method based on data analysis provided by the embodiment one of the present application, as shown in the figure, the method can include: Figure 1 Step S101, obtaining the oil temperature data of each monitoring point on the frying production line at each time in a preset number of historical periods, obtaining the temperature data sequence of each monitoring point in each historical period, the equipment on the frying production line in the historical period has no alarm record, and the frying products are all qualified.
[0020] Industry usually sets temperature sensors at each key point of the frying production line to collect the oil temperature data at each key point, and then introduces PCA-based multivariate statistical process monitoring (MSPM) technology to monitor the fault of the frying production line, that is, according to the oil temperature data at each key point, the control limit of T² statistic and SPE statistic is calculated, and finally the fault of the frying production line is monitored according to the control limit of T² statistic and SPE statistic.
[0021] Since the control limit of T² statistic and SPE statistic calculated completely depends on the oil temperature data at each key point, the quality of the oil temperature data at each key point determines the quality of the control limit of T² statistic and SPE statistic, and further determines the accuracy of the fault monitoring of the frying production line. Among the oil temperature data at each key point used to calculate the control limit of T² statistic and SPE statistic, there may be unobserved early fault data, such as local oil temperature anomaly, which is a typical precursor of fault, usually showing that the oil temperature at a specific monitoring point has a slow drift, or the difference with the neighborhood oil temperature has a gradually increasing trend. This fault is formed insidiously, if high-efficiency and intelligent monitoring means is not adopted, when it develops into obvious faults such as serious coking of heating pipe and deterioration of oil quality, it will cause serious economic losses such as product batch rejection, energy consumption surge and equipment damage.
[0022] Therefore, in the embodiment of the present application, the collected oil temperature data at each key point is purified to select the healthiest data for calculating the control limit of T² statistic and SPE statistic, so as to realize more accurate perception of the potential fault existing on the frying production line, and further improve the accuracy and sensitivity of the fault monitoring of the frying production line.
[0023] Firstly, high-precision temperature sensors (such as NTC temperature sensors) are deployed at each monitoring point of the frying production line, and the collection frequency of the temperature sensors is set to 1Hz, at the same time, from the start of heating to the end of heating is regarded as an oil frying production period, the oil temperature data of each monitoring point at each time in each oil frying production period is recorded, and the quality inspection result of the frying products in each oil frying production period is recorded.
[0024] In all the oil frying production cycles, the equipment on the oil frying production line has no alarm record, and the oil frying products are all qualified, and the oil frying production cycle is recorded as a history cycle, and the oil temperature data in the history cycle is considered to be normal data in a macroscopic view. In order to ensure that the analysis result has sufficient statistical efficiency, the number of history cycles needs to be ensured to be not less than 30, that is, the preset number is set to 30, which is not limited here, and the implementer can set it according to the specific scene. Although the oil temperature data in the history cycle is normal data in a macroscopic view, there may still be micro abnormal data of early faults that are not detected, so the temperature data sequence composed of the oil temperature data of each monitoring point in each history cycle is used to analyze and purify all the temperature data sequences, and the healthiest data is selected to calculate the control limit of the T2 statistic and the SPE statistic.
[0025] In step S102, according to the data fluctuation of the temperature data sequence of each monitoring point in each history cycle, the possible healthy monitoring points in each history cycle are obtained, and the history cycle in which all the monitoring points are possible healthy monitoring points is recorded as a possible healthy cycle.
[0026] Since the fault often appears as an abnormal single monitoring point in the oil frying production process, in the embodiment of the present application, first, the data quality at each monitoring point in each history cycle is analyzed according to the data fluctuation of the temperature data sequence of each monitoring point in each history cycle, so as to classify all the monitoring points in each history cycle into two categories, to obtain the possible healthy monitoring points in each history cycle, and then the history cycle in which all the monitoring points are possible healthy monitoring points is recorded as a possible healthy cycle, so as to realize the preliminary screening of the oil temperature data in all the history cycles.
[0027] Taking the tth history cycle as an example, the steps of obtaining the possible healthy monitoring points in the tth history cycle are as follows: (1) Taking the ith monitoring point as an example, according to the heating rate of the ith monitoring point in the heating period and the fluctuation of the oil temperature data of the ith monitoring point in the holding period in the tth history cycle, the first abnormality degree of the ith monitoring point in the tth history cycle is obtained.
[0028] In a complete oil frying production cycle, the heating period and the holding period are included, the heating period is the period from starting heating to the oil temperature reaching the target temperature (i.e. the required temperature of the oil frying product), which is used to heat the oil temperature to the required temperature of the oil frying product to ensure the quality of the oil frying product; and the holding period is the period from the oil temperature in the oil frying pot reaching the target temperature to the end of heating, which is used to stabilize the oil temperature in the oil frying pot at the target temperature (there is an allowable fluctuation range, such as ±2°C), so as to ensure that the oil frying conditions remain consistent when new oil frying products are put in, and the stability of the product quality is ensured.
[0029] Since the temperature maintaining stage usually adopts advanced PID controller and "dead zone" control to realize accurate temperature maintaining, potential problems of the equipment are most likely to be exposed in the temperature maintaining stage, such as temperature sensor drift anomaly, insufficient power of heating tube, PID controller parameter disorder, etc., so the first abnormality degree of the i th monitoring point in the t th historical period is obtained according to the heating rate of the i th monitoring point in the heating period and the fluctuation of the oil temperature data in the temperature maintaining period in the t th historical period, to reflect the health condition of the i th monitoring point itself, specifically: obtaining the heating rate of the i th monitoring point in the t th historical period according to the oil temperature data of the i th monitoring point in the heating period in the t th historical period; obtaining the theoretical heating rate at the i th monitoring point, calculating the absolute value of the difference between the heating rate and the theoretical heating rate to obtain a heating rate deviation value, taking the heating rate deviation value as the numerator and the theoretical heating rate as the denominator to obtain a heating rate deviation degree of the i th monitoring point, linearly normalizing the heating rate deviation degree to obtain a heating rate deviation degree normalization value; taking the oil temperature data of the i th monitoring point in the temperature maintaining period in the t th historical period as temperature maintaining data, calculating the coefficient of variation of all temperature maintaining data, linearly normalizing the coefficient of variation to obtain a coefficient of variation normalization value; weighting and summing the heating rate deviation degree normalization value and the coefficient of variation normalization value to obtain the first abnormality degree of the i th monitoring point in the t th historical period.
[0030] In an embodiment, the calculation formula of the first abnormality degree of the i th monitoring point in the t th historical period is:
[0031] wherein, represents the first abnormality degree of the i th monitoring point in the t th historical period, represents the heating rate of the i th monitoring point in the t th historical period, represents the theoretical heating rate at the i th monitoring point, represents the standard deviation of all oil temperature data of the i th monitoring point in the temperature maintaining period in the t th historical period, represents the average value of all oil temperature data of the i th monitoring point in the temperature maintaining period in the t th historical period, which is the coefficient of variation, represents the first weight, represents the second weight, represents a linear normalization function, represents the absolute value of the difference.
[0032] It should be noted that, The greater the value is, the greater the degree of deviation of the heating rate of the heating period of the i-th monitoring point in the t-th historical period is, and further The greater the value is, the more likely the i-th monitoring point has a potential fault in the t-th historical period is; The greater the value is, the more unstable the oil temperature at the i-th monitoring point in the holding period of the t-th historical period is, and further The greater the value is, the more likely the i-th monitoring point has a potential fault in the t-th historical period is. Since the heating period and the holding period are equally important in data quality assessment, the value is set to be Here, no limitation is made, and the implementer can set it according to the specific scene.
[0033] (2) According to the overall fluctuation of the temperature data sequence of the i-th monitoring point in the t-th historical period, a second abnormality degree of the i-th monitoring point in the t-th historical period is obtained.
[0034] Specifically: linear fitting is performed on the temperature data sequence of the i-th monitoring point in the t-th historical period to obtain a fitting straight line, and the absolute value of the slope of the fitting straight line is linearly normalized to obtain an oil temperature fluctuation trend value of the i-th monitoring point in the t-th historical period; The standard deviation of all data in the temperature data sequence of the i-th monitoring point in the t-th historical period is obtained, and the standard deviation is linearly normalized to obtain an oil temperature fluctuation intensity value of the i-th monitoring point in the t-th historical period; The mean value between the oil temperature fluctuation trend value and the oil temperature fluctuation intensity value is calculated to obtain the second abnormality degree of the i-th monitoring point in the t-th historical period.
[0035] In an embodiment, the calculation formula of the second abnormality degree of the i-th monitoring point in the t-th historical period is:
[0036] Wherein, represents the second abnormality degree of the i-th monitoring point in the t-th historical period, represents the standard deviation of all data in the temperature data sequence of the i-th monitoring point in the t-th historical period, represents the absolute value of the slope of the fitting straight line corresponding to the temperature data sequence of the i-th monitoring point in the t-th historical period, represents a linear normalization function.
[0037] It should be noted that, in the process of frying production, in order to ensure that the frying production line is 100% continuous feeding, the heat preservation period will be longer, for example, in 24 hours a day, there may be only half an hour to two hours in the heating period, and the remaining more than 20 hours are heat preservation period, and in the ideal state, the oil temperature in the heat preservation period remains stable, so the slope of the fitting straight line corresponding to the temperature data sequence of the i th monitoring point in the t th historical period will be close to 0, The greater, the more likely it is that the oil temperature data of the i th monitoring point in the t th historical period has a trend anomaly, and further The greater, the more likely it is that the i th monitoring point has a potential failure in the t th historical period; The greater, the greater the overall fluctuation intensity of the oil temperature data of the i th monitoring point in the t th historical period, and further The greater, the more likely it is that the i th monitoring point has a potential failure in the t th historical period.
[0038] (3) According to the first abnormality degree and the second abnormality degree of the i th monitoring point in the t th historical period, the weighted abnormality index of the i th monitoring point in the t th historical period is obtained.
[0039] The first abnormality degree of the i th monitoring point in the t th historical period is taken as the core index, and the second abnormality degree of the i th monitoring point in the t th historical period is taken as the important supplementary verification index, and the first abnormality degree and the second abnormality degree are weighted and summed to obtain the weighted abnormality index of the i th monitoring point in the t th historical period.
[0040] In an embodiment, the calculation formula of the weighted abnormality index of the i th monitoring point in the t th historical period is:
[0041] Wherein, The weighted abnormality index of the i th monitoring point in the t th historical period is represented by, The first abnormality degree of the i th monitoring point in the t th historical period is represented by, The second abnormality degree of the i th monitoring point in the t th historical period is represented by, The first abnormality degree corresponds to the weight, The second abnormality degree corresponds to the weight.
[0042] It should be noted that, The greater, The greater, the more likely it is that the i th monitoring point has an unnoticed potential failure in the t th historical period, and further The greater; since the first abnormality degree is the core index and the second abnormality degree is the supplementary verification index, the first abnormality degree A higher weight is given to the second degree of anomaly. In this embodiment of the invention, a lower weight is set. , There are no restrictions here; implementers can set them according to the specific scenario.
[0043] (4) Obtain the weighted abnormality index of each monitoring point in the t-th historical period, and obtain the possible health monitoring points in the t-th historical period based on the weighted abnormality index of each monitoring point in the t-th historical period.
[0044] Specifically: the maximum value of the weighted anomaly index of all monitoring points within any given historical period is selected and recorded as the maximum weighted anomaly index of that given historical period; Obtain the maximum weighted anomaly index for each historical period, and use the 99th percentile of all maximum weighted anomaly indices as the anomaly index threshold. There is no restriction here, and implementers can set it according to specific scenarios. If the weighted anomaly index of the i-th monitoring point in the t-th historical period is less than the anomaly index threshold, then the i-th monitoring point is recorded as a possible health monitoring point in the t-th historical period. Similarly, all possible health monitoring points in the t-th historical period are obtained.
[0045] Following the method described above for obtaining all possible health monitoring points in the t-th historical period, all possible health monitoring points in each historical period are obtained. The historical period in which all monitoring points are possible health monitoring points is recorded as a possible health period. The quality of oil temperature data at each monitoring point in a possible health period is relatively healthy.
[0046] Step S103: For any possible healthy cycle, based on the difference between the oil temperature data of all monitoring points in each time window of the possible healthy cycle and the oil temperature data of all monitoring points in all historical cycles, obtain the comprehensive deviation index of the possible healthy cycle. Based on the comprehensive deviation index of each possible healthy cycle, purify all temperature data sequences to obtain the final health data.
[0047] During the frying process, there are inherent correlations between various monitoring points, determined by the physical structure of the equipment and the oil flow circulation pattern. For example, monitoring points close to the heating source should have a high temperature correlation, while monitoring points in symmetrical positions should have similar temperature change patterns. Even if the temperature sensor readings at a single monitoring point are normal, if this cooperative relationship between monitoring points is disrupted, it may indicate a system-level failure (such as abnormal oil flow circulation or uneven heating).
[0048] The possible health period obtained in step S102 is only based on a single monitoring point, can only exclude part of the obviously unhealthy data, does not consider the cooperative relationship between multiple monitoring points, cannot exclude unhealthy data whose cooperative relationship is broken, and therefore needs to further purify the data in the possible health period to obtain final health data, so as to screen the most healthy data for calculating the control limit of the T2 statistic and the SPE statistic, thereby realizing more accurate perception of potential faults existing on the frying production line and further improving the accuracy and sensitivity of fault monitoring of the frying production line.
[0049] In the multivariate statistical process monitoring (MSPM) technology based on PCA, the first step is to construct a PCA model, and the first step of constructing the PCA model is to calculate the covariance matrix of the temperature data sequence of all monitoring points in all health periods. The cooperative relationship between two monitoring points is completely described in the covariance matrix, and therefore in the embodiment of the application, the cooperative relationship between multiple monitoring points is quantified according to the covariance matrix.
[0050] Taking the u-th possible health period as an example, the u-th possible health period is divided into at least two time windows, for example, each time window is 10 minutes, which is not limited here, and the implementer can dynamically set according to the cycle length of the frying production cycle. The covariance matrix is constructed according to the temperature data sequence of all monitoring points in each time window, and the local time covariance matrix of each time window is obtained. At the same time, the covariance matrix is constructed according to the temperature data sequence of all monitoring points in all historical periods, which is denoted as a reference covariance matrix. The construction of the covariance matrix is a prior art, and will not be described here.
[0051] Since all the equipment on the frying production line in all historical periods has no alarm faults, and the frying products are all qualified, that is, the data in the temperature data sequence of all monitoring points in all historical periods are normal data in the whole, the cooperative relationship described by the reference covariance matrix can be used as a healthy cooperative relationship. If the data in a certain time window is abnormal, the cooperative relationship described by the local time covariance matrix of the time window will deviate from the healthy cooperative relationship. Therefore, the comprehensive deviation index of the u-th possible health period can be obtained according to the difference between the local time covariance matrix of each time window in the u-th possible health period and the reference covariance matrix, so as to effectively capture the unhealthy data of systematic cooperative abnormality which is difficult to find in single-point analysis but will seriously affect the performance of the PCA model.
[0052] The step of obtaining the comprehensive deviation index of the u-th possible health period is: (1) Taking the wth time window in the u th possible health period as an example, the Frobenius norm between the local time covariance matrix of the wth time window and the reference covariance matrix is calculated to obtain the degree of cooperative abnormality of all monitoring points in the wth time window in the u th possible health period. The greater the Frobenius norm, the greater the deviation between the cooperative relationship among all monitoring points in the wth time window and the healthy cooperative relationship, and the more likely there is a cooperative abnormality among all monitoring points in the wth time window in the u th possible health period, that is, in the wth time window in the u th possible health period, the more likely there is a monitoring point that has a potential fault that has not been detected. The Frobenius norm is prior art, which will not be described here.
[0053] (2) Obtain the degree of cooperative abnormality of all monitoring points in each time window in the u th possible health period, and obtain the maximum value among all degrees of cooperative abnormality, denoted as the comprehensive deviation index of the u th possible health period.
[0054] Similarly, obtain the comprehensive deviation index of each possible health period, and record the 95% quantile of all comprehensive deviation indexes as the comprehensive deviation index threshold. This is not limited here, and the implementer can set it according to the specific scene. The possible health period with a comprehensive deviation index less than the comprehensive deviation index threshold is recorded as the final health period, and the data in the temperature data sequence of each monitoring point in each final health period is recorded as the final health data.
[0055] At this point, the purification of the temperature data sequence of all monitoring points in all historical periods is completed, and the final health data is obtained.
[0056] Step S104, using PCA, obtain the control limit of T 2 statistics and SPE statistics of all final health data, obtain real-time T 2 statistics and real-time SPE statistics according to real-time oil temperature data of each monitoring point, and perform real-time fault monitoring on the frying production line according to the real-time T 2 statistics and real-time SPE statistics, and the control limit of T 2 statistics and SPE statistics.
[0057] After obtaining the final health data through step S103, the fault of the frying production line is monitored by using the PCA-based multivariate statistical process monitoring technology, and the general process is as follows: (1) data preprocessing, standardizing all final health data; (2) establishing a PCA model and calculating control limits, constructing a PCA model according to the standardized final health data, using the constructed PCA model to obtain the control limits of the T 2 statistics and the SPE statistics of all final health data; (3) online monitoring, real-time oil temperature data of each monitoring point on the frying production line is obtained, and the real-time T 2 statistics and the real-time SPE statistics of all real-time oil temperature data are obtained by using the above-constructed PCA model, and the real-time T 2 statistics and the real-time SPE statistics are compared with the control limits of the T 2 statistics and the SPE statistics; (4) fault diagnosis, if the real-time T 2 statistics does not exceed the control limit of the T 2 statistics, and the real-time SPE statistics does not exceed the control limit of the SPE statistics, it is determined that the frying production line does not have a fault; if the real-time T 2 statistics exceeds the control limit of the T 2 statistics, or the real-time SPE statistics exceeds the control limit of the SPE statistics, the real-time T 2 contribution map and the real-time SPE contribution map are generated according to the real-time oil temperature data of each monitoring point, the fault monitoring point in the frying production line is located in real time according to the real-time T 2 contribution map and the real-time SPE contribution map, and prompt information is sent to timely remind the relevant operator. The PCA-based multivariate statistical process monitoring technology is prior art, which will not be described here.
[0058] In summary, by analyzing the temperature data sequence of each monitoring point on the frying production line in each historical period, the possible health monitoring points in each historical period are obtained, and the historical period in which all monitoring points are possible health monitoring points is recorded as a possible health period, so as to exclude obviously unhealthy data, realize the preliminary screening of the temperature data sequence of all monitoring points in all historical periods, analyze the synergistic relationship between all monitoring points in each time window according to the difference between the oil temperature data of all monitoring points in each time window and the oil temperature data of all monitoring points in all historical periods, so as to further exclude unhealthy data whose synergistic relationship is broken, obtain final health data, complete the purification operation of the temperature data sequence of all monitoring points in all historical periods, and then use PCA to obtain the control limits of the T 2 statistics and the SPE statistics of all final health data, so as to monitor the real-time fault of the frying production line, and improve the accuracy and sensitivity of the fault monitoring of the frying production line.
[0059] The above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.
Claims
1. A data analysis-based fault monitoring method for frying production lines, characterized in that, The data analysis-based fault monitoring method for frying production lines includes: The oil temperature data of each monitoring point on the frying production line is obtained at each moment within a preset number of historical periods, and the temperature data sequence of each monitoring point within each historical period is obtained. During the historical period, there are no alarm records on the equipment of the frying production line, and the fried products are all qualified. Based on the data fluctuation of the temperature data sequence of each monitoring point in each historical period, the possible health monitoring points in each historical period are obtained, and the historical period in which all monitoring points are possible health monitoring points is recorded as the possible health period. For any possible healthy cycle, based on the difference between the oil temperature data of all monitoring points in each time window of the possible healthy cycle and the oil temperature data of all monitoring points in all historical cycles, the comprehensive deviation index of the possible healthy cycle is obtained. Based on the comprehensive deviation index of each possible healthy cycle, all temperature data sequences are purified to obtain the final health data. Using PCA, control limits for the T² and SPE statistics of all final health data are obtained. Based on the real-time oil temperature data of each monitoring point, real-time T² and SPE statistics are obtained. Based on the real-time T² and SPE statistics, as well as the control limits for the T² and SPE statistics, real-time fault monitoring of the frying production line is performed.
2. The data analysis-based fault monitoring method for a frying production line according to claim 1, characterized in that, The step of obtaining possible health monitoring points within each historical period based on the data fluctuation of the temperature data sequence of each monitoring point within each historical period includes: The historical period is divided into a heating period and a heat preservation period. For any historical period, the first degree of anomaly of any monitoring point in any historical period is obtained based on the heating rate of any monitoring point in the heating period and the fluctuation of oil temperature data in the heat preservation period. Based on the overall fluctuation of the temperature data sequence of any monitoring point within any historical period, the second degree of anomaly at any monitoring point within any historical period is obtained; The weighted sum of the first degree of abnormality and the second degree of abnormality is used to obtain the weighted abnormality index of any monitoring point in any historical period. Obtain the weighted anomaly index of each monitoring point within any historical period, and select the maximum value among the weighted anomaly indices of all monitoring points within any historical period, which is recorded as the maximum weighted anomaly index of any historical period. Obtain the maximum weighted anomaly index for each historical period, and use the first preset quantile of all maximum weighted anomaly indices as the anomaly index threshold. If the weighted anomaly index of any monitoring point within any historical period is less than the anomaly index threshold, then the monitoring point is recorded as a possible health monitoring point within the historical period.
3. The data analysis-based fault monitoring method for a frying production line according to claim 2, characterized in that, The step of obtaining the first degree of anomaly for any monitoring point within any historical period based on the heating rate during the heating period and the fluctuation of oil temperature data during the heat preservation period at any monitoring point within any historical period includes: Based on the oil temperature data of any monitoring point during the heating period in any historical cycle, obtain the heating rate of any monitoring point in any historical cycle. The theoretical heating rate at any of the monitoring points is obtained, the absolute value of the difference between the heating rate and the theoretical heating rate is calculated to obtain the heating rate deviation value, the heating rate deviation value is used as the numerator and the theoretical heating rate is used as the denominator to obtain the degree of heating rate deviation at any of the monitoring points, and the degree of heating rate deviation is linearly normalized to obtain the normalized value of the degree of heating rate deviation. The oil temperature data of any monitoring point during the heat preservation period of any historical cycle is recorded as heat preservation data. The coefficient of variation of all heat preservation data is calculated, and the coefficient of variation is linearly normalized to obtain the normalized value of the coefficient of variation. The weighted sum of the normalized value of the heating rate deviation and the normalized value of the coefficient of variation is used to obtain the first degree of anomaly at any monitoring point within any historical period.
4. The data analysis-based fault monitoring method for a frying production line according to claim 2, characterized in that, The step of obtaining the second anomaly level at any monitoring point within any historical period based on the overall fluctuation of the temperature data sequence of any monitoring point within any historical period includes: A linear fit is performed on the temperature data sequence of any monitoring point within any historical period to obtain a fitted straight line. The absolute value of the slope of the fitted straight line is linearly normalized to obtain the oil temperature fluctuation trend value of any monitoring point within any historical period. Obtain the standard deviation of all data in the temperature data sequence of any monitoring point within any historical period, and perform linear normalization on the standard deviation to obtain the oil temperature fluctuation intensity value at any monitoring point within any historical period. Calculate the average between the oil temperature fluctuation trend value and the oil temperature fluctuation intensity value to obtain the second anomaly level at any monitoring point within any historical period.
5. The data analysis-based fault monitoring method for a frying production line according to claim 1, characterized in that, The step of obtaining the comprehensive deviation index of any possible healthy cycle based on the difference between the oil temperature data of all monitoring points within each time window of any possible healthy cycle and the oil temperature data of all monitoring points within all historical cycles includes: Each possible health cycle is divided into at least two time windows. For any time window, a covariance matrix is constructed based on the oil temperature data of each monitoring point within the time window, which is denoted as the local time covariance matrix. A covariance matrix is constructed based on the temperature data sequences of all monitoring points in all historical periods, denoted as the reference covariance matrix. Calculate the Frobenius norm between the local time covariance matrix and the reference covariance matrix to obtain the degree of coordinated anomaly of all monitoring points in any possible health cycle within any time window; Obtain the degree of coordinated anomaly of all monitoring points in each time window within any possible healthy cycle, and obtain the maximum value among all coordinated anomaly degrees, which is recorded as the comprehensive deviation index of any possible healthy cycle.
6. The data analysis-based fault monitoring method for a frying production line according to claim 1, characterized in that, The process involves refining all temperature data sequences based on the comprehensive deviation index of each possible healthy cycle to obtain the final health data, including: Obtain the second preset quantile of the comprehensive deviation index of all possible health cycles, and denot it as the comprehensive deviation index threshold. Denote the possible health cycles with a comprehensive deviation index less than the comprehensive deviation index threshold as the final health cycles. Denote the data in the temperature data sequence of each monitoring point within each final health cycle as the final health data.
7. The data analysis-based fault monitoring method for a frying production line according to claim 1, characterized in that, The real-time fault monitoring of the frying production line based on real-time T² statistics, real-time SPE statistics, and control limits for T² statistics and SPE statistics includes: If the real-time T² statistic does not exceed the control limit of the T² statistic and the real-time SPE statistic does not exceed the control limit of the SPE statistic, then it is determined that the frying production line has not malfunctioned. If the real-time T² statistic exceeds the control limit of the T² statistic, or the real-time SPE statistic exceeds the control limit of the SPE statistic, then a real-time T² contribution map and a real-time SPE contribution map are generated based on the real-time oil temperature data of each monitoring point. The fault monitoring point in the frying production line is located in real time based on the real-time T² contribution map and the real-time SPE contribution map.