A Deep Learning Method and Device for CIP Process Parameters of a Beverage Production Line

Abstract

The present invention discloses a deep learning method and device for CIP process parameters of a beverage production line. The method includes collecting production process data of CIP process steps; reducing the scale of the CIP process data set; calculating the Pearson coefficients of various factors in the production process data; finding the optimal process parameter combination through the results of orthogonal experiments; normalizing the above process parameter combination results, and training the optimal process parameter combination through a neural network model to achieve a better cleaning solution, solving the problems that the existing CIP cleaning process parameters mainly rely on quality inspection and manual experience for static setting, with poor pertinence in the actual cleaning process, resulting in problems such as excessive CIP cleaning, low CIP efficiency, and waste of acid-base solution consumption. On the premise of ensuring the cleaning effect, it saves cleaning time and improves the efficiency of CIP cleaning.

Description

Technical Field

[0001] The present invention relates to the field of food, and particularly to a CIP (Clean in Place) cleaning process. In particular, it is suitable for a deep learning method and device for CIP process parameters of a beverage production line. Background Art

[0002] CIP refers to the use of turbulent cleaning liquid in a closed-loop system to circulate and clean various equipment in the production system such as tanks, pipelines, pumps, and filling machines without disassembling or moving the production equipment; by controlling conditions such as the concentration, temperature, cleaning time, pressure, and flow rate of the cleaning liquid, rapid and efficient cleaning of equipment and pipelines can be achieved.

[0003] Traditional CIP generally sets cleaning process parameters through verification of cleaning effects during the process planning stage. For example, in the "CIP cleaning system" disclosed in the Chinese patent literature with the publication number CN114769230A, the conductivity of the return water in the production equipment is used as a judgment criterion, and the cleaning effect is determined by manual judgment, so as to set the CIP cleaning process parameters. As a method of the conventional CIP cleaning process, this method still has problems such as large cleaning losses, too long cleaning time, and poor cleaning quality caused by the inability to set process parameters specifically. Summary of the Invention

[0004] The present invention discloses a deep learning method and device for CIP process parameters of a beverage production line. The method includes collecting production process data of CIP process steps; reducing the scale of the CIP process data set; calculating the Pearson coefficients of various factors in the production process data; finding the optimal process parameter combination through the results of orthogonal experiments; normalizing the above process parameter combination results, and training the optimal process parameter combination through a neural network model to achieve a better cleaning plan, solving the problems that the existing CIP cleaning process parameters mainly rely on quality inspection and manual experience for static setting, with poor pertinence in the actual cleaning process, resulting in problems such as excessive CIP cleaning, low CIP efficiency, and waste of acid and alkali solution consumption. On the premise of ensuring the cleaning effect, the cleaning time is saved and the efficiency of CIP cleaning is improved.

[0005] In order to achieve the above invention purpose, the present invention provides the following technical solutions.

[0006] A deep learning method and device for CIP process parameters, the method steps include:

[0007] S1 Collect the production process data of the CIP process steps. CIP is generally divided into three-step washing (pre-washing with water, alkali washing, post-washing with water) and five-step washing (pre-washing with water, alkali washing, intermediate washing with water, acid washing, post-washing with water) according to the process steps. The CIP production process data is mainly divided into three categories: one is the process data of each CIP process step; the second is the basic data of the CIP system; the third is the process data of the CIP object.

[0008] The process data of each CIP step mainly includes: production process data such as start time, end time, start temperature, end temperature, start flow rate, end flow rate, start reflux temperature, end reflux temperature, start reflux conductivity, and end reflux conductivity.

[0009] The basic data of CIP mainly includes: the type of CIP cleaning agent, the structural characteristics of the CIP cleaning object (such as the pipeline plane structure), the material of the CIP cleaning object, etc.

[0010] The process data of the CIP cleaning object mainly includes: the CIP object (such as the production equipment name and number), the production plan of the production equipment (such as product name, output, production time consumption), the CIP sharing status, etc.

[0011] S2 The present invention proposes a data acquisition and preprocessing method for CIP process data. This method is mainly used to reduce the scale of the CIP process data set through feature extraction and orthogonal analysis on the premise of ensuring the effectiveness of data analysis. Taking the various data in S1 as the object, a mixed-level orthogonal analysis scheme such as L64(47×86) is adopted, combined with the selected factor levels, that is, the number of each value of each factor in terms of quantity or quality in the whole scheme, to calculate the Pearson coefficient of each factor. On this basis, a corresponding data acquisition scheme is designed.

[0012] S3 The present invention proposes a CIP process model based on a deep learning model. This model is based on S1 and S2, and uses the deep learning method to establish a non-linear relationship model between CIP time, acid-base consumption and the main influencing factors of CIP. This method is mainly to improve the learning ability, generalization ability and model accuracy of the model for sampled data. Since the mean square error is the standard for measuring the stability of the model, the data set obtained by the S2 mixed orthogonal analysis is divided into a test set and a training set.

[0013] Preferably, different initial weights and biases are set for the training set data. After each training, the mean square error of the model in the test set is automatically calculated and compared with the mean square error of the previous model. The network with the smaller mean square error is retained. After a certain number of cyclic trainings, a neural network with the smallest mean square error is generated. The start time, end time, start temperature, end temperature, start flow rate, end flow rate, types of products produced by the CIP object (such as a filling machine), CIP co-line conditions, structural characteristics of the CIP object, etc. are used as the input quantities of the neural network, and the CIP duration and acid-base consumption are used as the output quantities to establish a neural network model.

[0014] After the neural network model is established through S3 in S4, a set of process parameter combinations (a set of neural network input parameter combinations) with the smallest given CIP duration and acid-base consumption is found through the orthogonal test results. The grey wolf algorithm (GWO) is used to find the optimal combination of process parameters to generate the smallest CIP duration and acid-base consumption and improve the efficiency of the CIP process.

[0015] Preferably, the results of the above process parameter combinations are normalized. Considering that the CIP duration and acid-base consumption are equally important, the weighting factors are the same, each being 1 / 2. The upper and lower limits of the grey wolf search are set to prevent out-of-bounds search. Combining with the neural network model, the optimal process parameter combination is trained to achieve a process parameter combination better than the orthogonal test.

[0016] The present invention proposes an edge computing device for optimizing CIP process parameters. This device is mainly used for the auxiliary decision-making of real-time control in the CIP process to realize the dynamic adjustment and control of the CIP process parameters. Based on the data communication acquisition device, the CIP process data is collected, the CIP process auxiliary decision-making algorithm is embedded, the optimal process parameters of the CIP cleaning process are calculated in real time, and the calculation results are uploaded to the MES service platform through the data communication acquisition device.

[0017] The edge computing device has a fast response speed and calculation speed, and can adapt to load changes and embed various intelligent algorithms to realize the dynamic decision-making of the process parameters in the CIP cleaning process.

[0018] The data communication acquisition device is used for collecting CIP analysis-related data and is deployed on the device side of the CIP system. The data collected by the device includes three types of data. One is to collect data from the production equipment control units PCS or SCADA such as the dispensing tank and filling machine, the second is for the CIP system to collect data through the device control unit, and the third is to collect the operation data of the manufacturing execution system MES in real time. The data communication acquisition collects the data and sends it to the edge computing device at a specified frequency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the deep learning method for CIP process parameters of the present invention.

[0020] Figure 2 It is a flowchart of the method proposed in the embodiment of the present invention.

[0021] Figure 3 It is a flowchart of the algorithm of the present invention.

[0022] Figure 4 It is a flowchart of the device data exchange of the CIP cleaning process.

[0023] Figure 5 It is a schematic diagram of the working process structure of the edge computing device of the present invention. Specific implementation manners

[0024] By providing a deep learning method and device for CIP process parameters of a beverage production line in the embodiments of the present application, the problems of poor pertinence, over-cleaning, and waste of cleaning liquid consumption in the CIP cleaning process in the production line in the prior art are solved, and the technical effect of providing corresponding CIP auxiliary decisions for different CIP production scenarios through data processing and data modeling, saving cleaning time, and improving the CIP cleaning efficiency on the premise of ensuring the cleaning effect is achieved.

[0025] The principle of the method is as Figure 1 shown. The method includes collecting production process data of CIP process steps; reducing the scale of the CIP process data set; calculating the Pearson coefficients of various factors in the production process data; finding the optimal process parameter combination through the results of orthogonal experiments; normalizing the above process parameter combination results, and training the optimal process parameter combination through a neural network model to achieve a better cleaning plan.

[0026] Next, a preferred solution of a deep learning method and device for CIP process parameters of a beverage production line provided in Figures 2 to 5 this embodiment will be described in detail.

[0027] As Figure 2 shown, in a beverage production line, the acid, alkali, and hot water used for cleaning are placed in a storage tank, and a liquid level gauge, a conductivity meter, a PLC controller, an edge computing device, and a data acquisition device in the device are installed on both sides of the CIP production line cleaning device.

[0028] Before cleaning, confirm whether the equipment to be cleaned has been emptied. If not, open the valve to empty the residual liquid in the equipment.

[0029] Perform the cleaning steps of alkali cleaning, water cleaning, acid cleaning, and water cleaning. Under the action of a circulation pump, the reusable acid liquid and alkali liquid are refluxed.

[0030] Collect the production process data of the CIP process steps, including the start time, end time, start temperature, end temperature, start flow rate, end flow rate, start reflux temperature, end reflux temperature, start reflux conductivity, end reflux conductivity, etc.

[0031] Propose a data acquisition and preprocessing method for CIP process data. This method is mainly used to reduce the scale of the CIP process data set through feature extraction and orthogonal analysis on the premise of ensuring effective data analysis. Taking the various data mentioned above as objects, adopt a mixed-level orthogonal analysis scheme, such as L64(47×86), combine with the selected factor levels, calculate the Pearson coefficient of each factor, and on this basis, design and form the corresponding data acquisition scheme.

[0032] Adopt the deep learning method to establish a neural network model for the CIP process; this model is based on the above, and adopts the deep learning method to establish a non-linear relationship model between CIP time, acid-base consumption and the main influencing factors of CIP.

[0033] This method is mainly to improve the learning ability, generalization ability and model accuracy of the model for the sampled data. Since the mean square error is the standard for measuring the stability of the model, the data set obtained by the above-mentioned mixed orthogonal analysis is divided into two parts: the test set and the training set.

[0034] Set different initial weights and biases for the training set data, automatically calculate the mean square error of the model in the test set after each training, and compare it with the mean square error of the previous model, and retain the network with the smaller mean square error. After a certain number of cyclic trainings, generate a neural network with the smallest mean square error.

[0035] Take the start time, end time, start temperature, end temperature, start flow rate, end flow rate, types of products produced by the CIP object (such as filling machine), CIP co-line working conditions, structural characteristics of the CIP object, etc. as the input variables of the neural network, and take the CIP duration and acid-base consumption as the output variables to establish a neural network model.

[0036] After establishing the neural network model through the above steps, find a set of process parameter combinations (a set of neural network input parameter combinations) with the smallest given CIP duration and acid-base consumption through the orthogonal test results, and adopt the Grey Wolf Optimization (GWO) algorithm to find the optimal combination of process parameters to generate the smallest CIP duration and acid-base consumption, and improve the efficiency of the CIP process.

[0037] Figure 3It is the flowchart of the Grey Wolf Algorithm. The optimization of the GWO algorithm starts with randomly creating a grey wolf population (candidate solutions). During the iteration process, the α, β, and δ wolves estimate the possible location of the prey (optimal solution). The grey wolves update their positions according to their distances from the prey. For exploration and exploitation during the search process, the parameter α should decrease from 2 to 0. If |A| > 1, the candidate solution is far from the prey; if |A| < 1, the candidate solution approaches the prey. The process of the GWO algorithm is as follows. The α-level wolf pack represents the leader in the population and is responsible for leading the entire wolf pack to hunt for prey, that is, the optimal solution in the optimization algorithm. The β-level wolf pack is responsible for assisting the α-level wolf pack, that is, the sub-optimal solution in the optimization algorithm. The δ-level wolf pack obeys the commands and decisions of α and β and is responsible for scouting, sentry duty, etc. The α and β with poor fitness will be demoted to δ. The ω-level wolf pack updates its position by surrounding α, β, or δ. |A| is used to simulate the attacking behavior of grey wolves towards the prey. In this embodiment, through the grey wolf algorithm and combined with the model obtained by deep learning, the optimal process parameters of the beverage production line for cleaning are calculated.

[0038] Normalize the results of the above process parameter combinations. Considering that the CIP duration, acid and alkali consumption are of equal importance, the weighting factors are the same, each being 1 / 2. Set the upper and lower limits of the grey wolf search to prevent the search from going out of bounds. Combine with the neural network model to train the optimal process parameter combination, achieving a process parameter combination better than that of the orthogonal experiment.

[0039] In addition to introducing the CIP cleaning method with in-depth learning, a deep learning device for the CIP process parameters of a beverage production line applied in this embodiment mainly includes a pure water tank, a hot water tank, an acid solution tank, an alkali solution tank, a concentrated alkali tank, a concentrated acid tank, a neutralization tank, a metering pump, a variable frequency circulating pump, a self-priming pump, a heat exchanger, a filter, a data communication and acquisition device, and an edge computing device. This device is mainly used for the auxiliary decision-making of the real-time control of the CIP process to realize the dynamic adjustment and control of the CIP process parameters.

[0040] Based on the data communication and acquisition device, collect the CIP process data, embed the CIP process auxiliary decision-making algorithm, calculate the optimal process parameters of the CIP cleaning process in real time, and upload the calculation results to the MES service platform through the data communication and acquisition device. Its working process is as Figure 4 shown.

[0041] The working process of the said edge computing device is as Figure 5As shown, real-time data is fused at the edge layer, and time-domain analysis, frequency-domain analysis, and PCA principal component analysis are performed during the fusion; the fused data is detected by a binary classification method; then diagnosis is performed by a multi-classification method; the data is uploaded to the MES service platform, and the fault diagnosis model and the novelty value detection model are trained through the data center of the MES service platform, which has a fast response speed and calculation speed, can adapt to load changes and embed various intelligent algorithms to achieve dynamic decision-making of the process parameters of the CIP cleaning process.

[0042] The data communication and acquisition device is used for acquiring CIP analysis-related data and is deployed on the device side of the CIP system. The data acquired by the device includes three types of data. One is to acquire data from the production equipment control units PCS or SCADA such as the dispensing tank and the filling machine. The second is that the CIP system acquires data through the device control unit. The third is to acquire the operation data of the production execution system MES in real time. The data is communicated and acquired and sent to the edge computing device at a specified frequency.

[0043] Due to the increasing diversity of products, the complexity of CIP objects such as production equipment is continuously increasing. Due to the differences in different products, different equipment, and different production processes, the corresponding CIP process parameters cannot be comprehensively and deeply verified for cleaning effects and quantitatively analyzed. Usually, for the production processes of different products and the CIP process parameters of production line equipment with different structures, process experience is mainly relied on, and the settings are relatively simple. Often, the same set of cleaning parameters is applied to both simple and complex cleaning objects, resulting in over-cleaning of simple cleaning objects.

[0044] Although the present application has been described in conjunction with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are considered to have covered any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A deep learning method for CIP process parameters of a beverage production line, characterized in that The method includes the following steps: S1: Collect the production process data of the CIP process steps, including: Process data of each CIP process step: including the time, temperature, flow rate, reflux temperature, and reflux conductivity at the start and end; Basic data of the CIP system: including the type of CIP cleaning agent, the structural characteristics of the CIP cleaning object, and the material of the CIP cleaning object; Process data of the CIP object: including the CIP object, the production plan of the production equipment, and the CIP sharing status; S2: Reduce the scale of the CIP process data set through feature extraction and mixed-level orthogonal analysis of the production process data; S3: Calculate the Pearson coefficients of each factor in the production process data, and thus design and form a corresponding data collection scheme; S4: Adopt a deep learning method to establish a neural network model of the CIP process; S5: Find a set of process parameter combinations with the minimum given CIP duration and acid-base consumption through the results of orthogonal experiments, and use the grey wolf algorithm to find the optimal process parameter combinations to generate the minimum CIP duration and acid-base consumption; S6: Normalize the above process parameter combination results, and through the neural network model, train the optimal process parameter combination to achieve a better cleaning scheme.

2. The deep learning method for CIP process parameters of a beverage production line according to claim 1, characterized in that, In S5, normalize the results of the smallest set of process parameter combinations, set the weighting factors of the CIP duration and acid-base consumption to 1 / 2, and set the upper and lower limits of grey wolf search.

3. A deep learning method for CIP process parameters of a beverage production line according to claim 1 or 2, characterized in that, S4 uses a deep learning method, takes the production process data of CIP as the input quantity of the neural network, and takes the CIP duration and acid-base consumption as the output quantity to establish a non-linear relationship model between CIP time, acid-base consumption, and CIP influencing factors.

4. A deep learning method for CIP process parameters of a beverage production line according to claim 1, characterized in that, The data set obtained by the mixed-level orthogonal analysis in S2 is divided into a test set and a training set.

5. A deep learning method for CIP process parameters of a beverage production line according to claim 4, characterized in that, Set different initial weights and biases for the training set data. After each training, automatically calculate the mean square error of the model in the test set and compare it with the mean square error of the previous model. Keep the network with the smaller mean square error. After a certain number of cyclic trainings, generate a neural network with the smallest mean square error.

6. An edge computing device for optimizing CIP process parameters of a beverage production line, applicable to the deep learning method according to any one of claims 1-5, characterized in that, The edge computing device is connected to the data communication and acquisition device, and the data communication and acquisition device is connected to the CIP controller.

7. The edge computing device for optimizing the CIP process parameters of a beverage production line according to claim 6, characterized in that, The edge computing device is connected to a SCADA real-time database based on the data communication and acquisition device and is connected to the MES database.

8. An edge computing device for optimizing CIP process parameters of a beverage production line according to claim 6, characterized in that, The output of the edge computing device is connected to the CIP controller.

Images