Dynamic analysis method for surface density of coating process

Abstract

The invention discloses a dynamic analysis method for surface density of a coating process, which comprises the following steps of: S1, acquiring real-time data, namely acquiring various data conditions in a coating process in real time; s2, entering an SPC analysis module, carrying out SPC statistical analysis on the obtained real-time data, and introducing CTP and CTQ setting modules in an output process; s3, a basic function of big data calculation is carried out, and after SPC statistical analysis is carried out, mean fluctuation changeepoi nt detection and interval feature extraction are carried out; and S4, the big data calculation application layer carries out judgment and analysis on the fluctuation trend, so that a diagnosis report is output, and the process is dynamically controlled after grading identification. According to the dynamic analysis method for the surface density of the coating process, on the basis of a conventional SPC, a trend detection module, a mean fluctuation detection module, a working condition sequence segmentation module, a sequence feature extraction module, a working condition clustering module, a TOP fluctuation sequence recognition module and other modules are introduced, and dynamic SPC management and control are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of coating process monitoring, and in particular to a surface density dynamic analysis method of a coating process. Background Art

[0002] The fluctuation of CTQ surface density in the coating process output affects product consistency, and CTQ is affected by the main CTP input characteristics, pump speed and GAP. It is particularly necessary to dynamically monitor process characteristic parameters based on real-time data and realize real-time SPC control. The current common SPC statistical method is based on the common control chart pattern recognition signal, which has the problem of single and rough recognition means, and insufficient analysis ability for non-stationary time series signals of industrial systems.

[0003] For example, the existing patent publication number is CN113832267, which discloses a blast furnace instrument data intelligent monitoring system and method based on SPC control theory, realizes blast furnace instrument monitoring based on conventional SPC XR control chart, and provides operation adjustment suggestions for different control modes.

[0004] For example, the existing patent publication number is CN111259334A, which discloses a method for monitoring and early warning of water consumption fluctuations of large industrial enterprises. It uses ks detection to determine whether it is normally distributed, uses XR control chart to realize SPC identification for normal distribution, and uses box plot to identify non-normal distribution.

[0005] For example, the existing patent publication number CN117113257A discloses a dynamic SPC control method based on a multi-value scenario. In order to solve the problem that conventional SPC statistics have fixed upper and lower limits and cannot be adjusted in real time, dynamic control is achieved by dynamically adjusting X(bar) based on rules.

[0006] The above patents all have shortcomings: conventional SPC XR control charts cannot realize trend and periodic signal recognition, and cannot statistically classify signals in real time. Summary of the invention

[0007] The purpose of the present invention is to provide a surface density dynamic analysis method for a coating process. On the basis of conventional SPC, trend detection, mean fluctuation detection and working condition sequence segmentation, sequence feature extraction and working condition clustering, TOP fluctuation sequence identification and other modules are introduced to realize dynamic SPC control.

[0008] The technical solution adopted by the surface density dynamic analysis method of a coating process disclosed in the present invention is:

[0009] A method for dynamic analysis of surface density of a coating process comprises the following steps:

[0010] S1, real-time data, real-time acquisition of various data during the coating process;

[0011] S2, enter the SPC analysis module, perform SPC statistical analysis on the acquired real-time data, and introduce the CTP and CTQ setting modules in the output process;

[0012] S3, the basic function of big data computing, performs mean fluctuation changepoint detection and interval feature extraction after SPC statistical analysis;

[0013] S4, the big data computing application layer, conducts judgment and analysis of fluctuation trends to output diagnostic reports and conducts dynamic control of post-grading and labeling processes.

[0014] As a preferred solution, the CTP and CTQ setting module uses CTQ surface density data as the main fluctuation factor of the current coating process, and the surface density is used as a product characteristic.

[0015] As a preferred solution, the SPC module mainly uses the X-BAR diagram and the control mode to determine whether the current signal exceeds the upper and lower limits, and realizes conventional fluctuation judgment analysis.

[0016] As a preferred solution, the SPC analysis module includes: tolerance judgment, SPC mean graph, out-of-bounds judgment and SPC control chart pattern recognition, the CTP and CTQ setting modules correspond to the SPC mean graph, and when the tolerance judgment and out-of-bounds judgment both exceed the set values, it is judged that the process is out of control.

[0017] As a preferred solution, the Changpoint is a common mean detection method, mainly used to identify the mean and variance jumps of time series. It takes the median filter trend signal as input, uses changepoint to realize mean fluctuation identification, and the output value is the time point corresponding to several jump points.

[0018] As a preferred solution, the Changepoint method includes the following:

[0019] 1. Statistical method: By constructing appropriate statistics to monitor the changes in the distribution characteristics of the data, calculate the statistics, and when the posterior probability exceeds the threshold, it is considered that there is a change point;

[0020] 2. Sliding window-based method: Slide a window of fixed length on the data sequence and compare the features of the data in the window. If the features between adjacent windows are very different, there may be a change point;

[0021] 3. Model-based approach: Use a time series model to fit the data and detect change points through changes in model parameters.

[0022] As a preferred solution, the feature extraction is based on the above changepoint detection. The original sequence is cut into several subsequences due to the change in mean value. The curve feature of each subsequence is realized by extracting mean, variance, slope, kurt, CP, and CPK.

[0023] As a preferred solution, the big data computing application layer includes interval feature extraction to determine whether there is a fluctuation trend. When the interval feature extraction shows a fluctuation trend, the fluctuation trend type is classified. After the fluctuation trend type is classified, fluctuation auxiliary positioning is performed.

[0024] As a preferred solution, the fluctuation trend type classification identifies the fluctuation changes in the output surface density and gives the fluctuation type. Based on the output fluctuation type, the qualitative impact of the current input on the output is located, and the impact fluctuation range and corresponding time point are given.

[0025] As a preferred solution, the fluctuation range assists in positioning: fluctuations in different surface densities represent changes in current working conditions and corresponding input changes. Based on the fluctuations in output surface density, the source of influence of input factors is reversely located, which can assist businesses in improving operations and reducing process fluctuations.

[0026] The beneficial effect of a surface density dynamic analysis method for a coating process disclosed in the present invention is: on the basis of conventional SPC, trend detection, mean fluctuation detection and working condition sequence segmentation, sequence feature extraction and working condition clustering, TOP fluctuation sequence identification and other modules are introduced to realize dynamic SPC control. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The present invention is a flow chart of a method for dynamically analyzing surface density of a coating process. DETAILED DESCRIPTION

[0028] The present invention will be further described and illustrated below in conjunction with specific embodiments and accompanying drawings:

[0029] Please refer to Figure 1 , a method for dynamic analysis of surface density of a coating process, comprising the following steps:

[0030] S1, real-time data, real-time acquisition of various data during the coating process;

[0031] S2, enter the SPC analysis module, perform SPC statistical analysis on the acquired real-time data, and introduce the CTP and CTQ setting modules in the output process.

[0032] The CTP and CTQ setting modules are mainly due to the CTQ surface density data being the main fluctuation factor in the current coating process, with surface density being a product characteristic.

[0033] SPC module: mainly uses X-BAR chart and control mode to determine whether the current signal exceeds the upper and lower limits, and realizes conventional fluctuation judgment analysis.

[0034] The SPC analysis module includes: tolerance judgment, SPC mean chart, out-of-bounds judgment and SPC control chart pattern recognition. The CTP and CTQ setting modules correspond to the SPC mean chart, and when the tolerance judgment and out-of-bounds judgment both exceed the set values, it is judged that the process is out of control.

[0035] S3, a basic function for big data computing, performs mean fluctuation changepoint detection and interval feature extraction after SPC statistical analysis.

[0036] Based on EMD (empirical mode), the trend, cycle and random signal of the surface density are decomposed to obtain the trend signal.

[0037] EMD decomposition principle:

[0038] EMD mainly solves the trend decomposition of nonlinear and non-stationary signals. The EMD decomposition calculation steps are as follows:

[0039] Finding extreme points: Obtain all the maximum and minimum values ​​of the signal sequence through the Find Peaks algorithm;

[0040] Fitting envelope curve: through the maximum and minimum value groups of the signal sequence, two smooth peak / trough fitting curves are obtained by cubic spline interpolation, namely the upper envelope and lower envelope of the signal;

[0041] Mean envelope: The mean envelope is obtained by averaging the two extreme value curves;

[0042] The intermediate signal is obtained by subtracting the mean envelope from the original signal;

[0043] Determine the intrinsic mode function (IMF). If it satisfies, continue to jump to ①, otherwise end;

[0044] After EMD extraction, the trend signal is retained after removing the non-periodic fluctuation signal, and the trend representation is more obvious.

[0045] Median filter:

[0046] At the same time, due to problems such as sampling accuracy and random disturbance, the extracted EMD trend signal has local fluctuations, and the trend is enhanced by using the median filter. The median filter signal is closer to the actual trend signal, and the median filter is obtained by filtering with a sliding window of n=10.

[0047] The median filter signal removes local fluctuation trends and is smoother while still showing the trend.

[0048] Changpoint is a common mean detection method, which is mainly used to identify the mean and variance jumps of time series. It takes the median filter trend signal as input, uses changepoint to realize mean fluctuation identification, and the output value is the time point corresponding to several jump points.

[0049] Changepoint methods include the following:

[0050] 1. Statistical method: By constructing appropriate statistics to monitor the changes in the distribution characteristics of the data, calculate the statistics, and when the posterior probability exceeds the threshold, it is considered that there is a change point;

[0051] 2. Sliding window-based method: Slide a window of fixed length on the data sequence and compare the features of the data in the window. If the features between adjacent windows are very different, there may be a change point;

[0052] 3. Model-based approach: Use a time series model to fit the data and detect change points through changes in model parameters.

[0053] Feature extraction is based on the above changepoint detection. The original sequence is cut into several subsequences due to the change in mean. The curve feature of each subsequence is realized by extracting mean, variance, slope, kurt, CP, and CPK.

[0054] Mean: Calculates the average value within a subsequence.

[0055] Variance: Calculates the within-subseries standard deviation.

[0056] Slope calculation:

[0057] Calculate the slope of change within the subsequence and evaluate different fluctuation trends with different slope sizes.

[0058] Kurt calculates:

[0059] Among them: kurtosis evaluates the kurtosis of the overall distribution curve in the current sequence, which is calculated by the fourth-order standard moment.

[0060] kurtosis = 0 - same steepness as normal distribution

[0061] kurtosis>0——steeper than the peak of normal distribution——spike peak

[0062] kurtosis<0——a platform higher than the peak of the normal distribution——a flat peak

[0063] CP calculation:

[0064] CP stands for process capability, which reflects the relative size of process fluctuation to specification limit, that is, process stability. Here, the difference between the upper and lower limit extremes in the current split subsequence is calculated.

[0065] Cp=(USL-LSL) / 6σ, USL and LSL are the upper and lower bounds of the subsequence respectively.

[0066] CPK calculation:

[0067] CPK reflects the degree of deviation between the process mean and the target. Here, the deviation between the upper and lower limit extreme values ​​and the mean in the current segmentation subsequence is calculated. CPK = min[(USL-mean), (mean-LSL)] / 3σ]

[0068] Mean is the mean of the subsequence, USL and LSL are the upper and lower bounds of the subsequence respectively.

[0069] S4, the big data computing application layer, conducts judgment and analysis of fluctuation trends to output diagnostic reports and conducts dynamic control of post-grading and labeling processes.

[0070] The big data computing application layer includes interval feature extraction to determine whether there is a fluctuation trend. When there is a fluctuation trend in interval feature extraction, the fluctuation trend type is classified. After the fluctuation trend type is classified, fluctuation auxiliary positioning is performed.

[0071] The fluctuation trend type classification identifies the fluctuation changes in the output surface density and gives the fluctuation type. Based on the output fluctuation type, it locates the qualitative impact of the current input on the output and gives the impact fluctuation range and corresponding time point.

[0072] Specifically: working condition clustering (surface density)

[0073] The surface density subsequences after sequence segmentation include ascending, descending and stable sequences. Different surface density fluctuations represent changes in the current working conditions and corresponding input changes.

[0074] Based on the above extracted features, surface density classification clustering is implemented based on GMM. Different fluctuation types are divided into corresponding types.

[0075] GMM clustering principle:

[0076] ① Superimpose multiple Gaussian models to form a mixed model, and use the mixed model to calculate the probability distribution of data

[0077] GMM k =λ1N(μ1,σ1)+...+λ k N(μ k , σ k )+...=∑λ k N(μk , σ k )

[0078] where λ k Indicates the probability that the current sample belongs to the kth distribution, μ k , σ k represents the mean and variance of the kth distribution

[0079] represents the probability density of the kth distribution

[0080] ②First initialize a set of parameters and Gaussian distribution categories, and use the EM expectation step to calculate λ k

[0081] ③ Use EM max step to calculate the kth distribution μ k , σ k

[0082] (number of data points of the kth Gaussian)

[0083] (The kth Gaussian accounts for the weight of the entire GMM model)

[0084] (mean of the kth Gaussian)

[0085] (variance of the kth Gaussian)

[0086] ④Iterate to step ② until EM converges, that is, calculate the Gaussian distribution coefficient λ of each sample k , and μ for each Gaussian distribution k , σ k ,, take λ=max(λ1, λ2,...,λ k ) is to calculate the Gaussian distribution category to which the current sample belongs.

[0087] Fluctuation range auxiliary positioning: Different surface density fluctuations represent changes in current working conditions and corresponding input changes. Based on the fluctuation of output surface density, the influencing source of input factors can be reversely located, which can assist businesses in improving operations and reducing process fluctuations.

[0088] The source of surface density fluctuation is:

[0089] ①④⑤:Tool changes

[0090] ③: Pump speed change

[0091] ②: Comprehensive changes in pump speed and tool.

[0092] The present invention provides a surface density dynamic analysis method for a coating process. On the basis of conventional SPC, trend detection, mean fluctuation detection and working condition sequence segmentation, sequence feature extraction and working condition clustering, TOP fluctuation sequence identification and other modules are introduced to realize dynamic SPC control.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for dynamic analysis of surface density of a coating process, characterized in that: The following steps are involved: S1, real-time data, real-time acquisition of various data during the coating process; S2, enter the SPC analysis module, perform SPC statistical analysis on the acquired real-time data, and introduce the CTP and CTQ setting modules in the output process; S3, the basic function of big data computing, performs mean fluctuation changepoint detection and interval feature extraction after SPC statistical analysis; S4, the big data computing application layer, conducts judgment and analysis of fluctuation trends to output diagnostic reports and conducts dynamic control of post-grading and labeling processes.

2. A method for dynamic analysis of surface density of a coating process according to claim 1, characterized in that: The CTP and CTQ setting module is because the main fluctuation factor of the current coating process is the CTQ surface density data, and the surface density is a product characteristic.

3. A method for dynamic analysis of surface density of a coating process according to claim 1, characterized in that: The SPC module mainly uses the X-BAR diagram and control mode to determine whether the current signal exceeds the upper and lower limits, and realizes conventional fluctuation judgment analysis.

4. A method for dynamic analysis of surface density of a coating process as claimed in claim 3, characterized in that: The SPC analysis module includes: tolerance judgment, SPC mean graph, out-of-bounds judgment and SPC control chart pattern recognition. The CTP and CTQ setting modules correspond to the SPC mean graph, and when the tolerance judgment and out-of-bounds judgment both exceed the set values, it is judged that the process is out of control.

5. A method for dynamic analysis of surface density of a coating process according to claim 1, characterized in that: The Changpoint is a common mean detection method, which is mainly used to identify the mean and variance jumps of time series. It takes the median filter trend signal as input, adopts changepoint to realize mean fluctuation identification, and the output value is the time point corresponding to several jump points.

6. A method for dynamic analysis of surface density of a coating process as claimed in claim 5, characterized in that: The Changepoint methods include the following:

1. Statistical method: By constructing appropriate statistics to monitor the changes in the distribution characteristics of the data, calculate the statistics, and when the posterior probability exceeds the threshold, it is considered that there is a change point; 2. Sliding window-based method: Slide a window of fixed length on the data sequence and compare the features of the data in the window. If the features between adjacent windows are very different, there may be a change point; 3. Model-based approach: Use a time series model to fit the data and detect change points through changes in model parameters.

7. A method for dynamic analysis of surface density of a coating process according to claim 1, characterized in that: The feature extraction is based on the above changepoint detection. The original sequence is cut into several subsequences due to the change in mean value. The curve feature of each subsequence is realized by extracting mean, variance, slope, kurt, CP, and CPK.

8. A method for dynamic analysis of surface density of a coating process according to claim 1, characterized in that: The big data computing application layer includes interval feature extraction to determine whether there is a fluctuation trend. When the interval feature extraction determines that there is a fluctuation trend, the fluctuation trend type is classified. After the fluctuation trend type is classified, fluctuation auxiliary positioning is performed.

9. A method for dynamic analysis of surface density of a coating process as claimed in claim 8, characterized in that: The fluctuation trend type classification identifies the fluctuation changes in the output surface density and gives the fluctuation type. Based on the output fluctuation type, the qualitative impact of the current input on the output is located, and the impact fluctuation range and corresponding time point are given.

10. A method for dynamic analysis of surface density of a coating process according to claim 8, characterized in that: The fluctuation range assists in positioning: fluctuations in different surface densities represent changes in current working conditions and corresponding input changes. Based on the fluctuations in output surface density, the source of influence of input factors is reversely positioned, which can assist businesses in improving operations and reducing process fluctuations.

Images