Analysis of automation technology education reform based on industrial internet and smart manufacturing promotion

The rapid development of China’s modern manufacturing industry in the context of industrial Internet and intelligent manufacturing has brought good development opportunities and higher requirements to the electrical automation technology program [1-2]. The industry needs more and more specialized technical talents, which have far exceeded the students currently enrolled within vocational college majors in China [3]. According to the investigation and analysis, the most lacking professional talents in industrial enterprises are divided into two kinds, one is professional design talents and the other is professional technical talents [4-5]. According to the analysis of enterprise demand, there is a certain demand for professional design talents, and professional and technical talents are in greater demand [6].

Colleges and universities, as an important position for training social talents, need to innovate their schooling concepts based on the job market in order to effectively improve the comprehensive quality, innovation, practical ability and social competitiveness of electrical automation technology students [7-9]. Colleges and universities should combine social development needs and market demand, make appropriate adjustments to the curriculum system and teaching content, constantly innovate and reform teaching methods, and then cultivate more composite electrical automation technology talents [10-11].

In colleges and universities, electrical automation technology is an emerging discipline with strong comprehensiveness. The specialty requires students to master electronic technology, computer network technology, microcomputer control technology and other basic knowledge, and master the working principle of electrical equipment, can use computer technology to realize the basic principles and methods of electrical control, but also need to have a certain degree of engineering design capabilities [12-14]. However, at present, in the teaching of professional courses of electrical automation technology in colleges and universities, most of the teachers are too focused on the explanation of the knowledge of the textbook, ignoring the cultivation of students’ practical ability. In order to meet the requirements of modern education, teachers in teaching not only need to strengthen students’ basic knowledge, but also need to focus on cultivating students’ practical ability, so that students can achieve a solid grasp of the basics at the same time, can be flexible to use the basic knowledge [15-17]. Moreover, with the increasing degree of automation in China’s industry, the development space of electrical automation technology has become more and more broad. In the context of the era, the direction of the professional curriculum reform needs to go deep into the production line of enterprises, combined with vocational education and the urgent needs of the industry, so that the curriculum teaching content can better meet the needs of society and in line with the development of science and technology [18-20].

Electrical automation teaching reform research is mainly divided into two directions, one is the improvement of teaching methods, and the other is the improvement of teaching technology. The research on the improvement of teaching methods of electrical automation is as follows: Xu, X introduced and analyzed the practical effect of “Excellence Plan 2.0” for electrical engineering and automation majors, and pointed out that the cultivation of high-quality electrical automation professionals requires the scientific and reasonable integration of cultivation objectives, curriculum system, faculty, and practice links [21]. Liu, D et al. proposed to improve the practical teaching reform of electrical engineering automation to improve the effect of practical teaching, and then improve the students’ practical application ability of electrical automation engineering [22]. And the research on the improvement of teaching technology of electrical automation has Xu, K Using the comparative teaching experiment method, it reveals that VR technology helps to stimulate students’ motivation and interest in learning electrical automation, and improves students’ classroom attendance [23]. Zou, J describes the introduction of advanced information technology into the electrical automation teaching classroom to replace the traditional teaching methods in order to better cultivate students’ innovative thinking and professional skills, and points out the necessity of transforming the lecture-based teaching mode to the student-centered teaching mode [24]. Huo, Y et al. conceptualized an automation teaching model with artificial intelligence and virtual simulation reality technology as the core logic, and improved the corresponding training system, which made a positive contribution to the cultivation of applied talents in electrical automation [25]. Cordeiro, A et al. envisioned an interactive educational program for industrial automation that could effectively motivate electrical automation engineering students to engage in professional problem solving and build a sound professional knowledge system [26]. All of the above studies have made important contributions to the reform of electrical automation education, however, none of these studies, have considered the actual environment and industrial real needs of the current industrial internet and smart manufacturing.

Based on the employment requirements for electrical automation majors generated under the promotion of industrial Internet and intelligent manufacturing, this paper realizes the teaching reform from various aspects, such as platformized curriculum system, strengthening the dynamic docking between electrical automation technology majors and industries, and constructing the cultivation mode combining on-campus practical training and enterprise internship. For the massive learning data generated, K-means clustering analysis algorithm is applied to classify students’ learning behavior into four major categories, which provides scientific guidance for students’ management. At the same time, the logistic regression model was applied to test the influence of four factors, namely, the selection of training bases, the training process, the internship process and the intelligent manufacturing facilities, on the students’ satisfaction. In this way, the improvement analysis of the teaching reform strategy of this paper is carried out.

In the blended teaching mode, the teacher’s evaluation of the learning effect of the technology platform knowledge points is a more complex issue, can not rely solely on the performance of electrical automation technology students to assess the teacher’s good or bad design of the knowledge points of the electrical automation technology professional learning. There are many factors that affect the learning effect of students, for example, students only watched the learning video once, without thinking and digesting to participate in the learning test related to electrical automation technology, students did not invest enough time in learning and did not pay attention to learning, electrical automation technology professional knowledge points of the organization of the design of the irrational and other factors will affect the evaluation of students’ performance.

In order to address the design evaluation problems in hybrid teaching reform, this paper proposes a K-means mean clustering algorithm [27]. The algorithm is a method for mining and analyzing the learning performance of students’ technology platform, clustering and grouping the learning effect of different knowledge points, and analyzing the final clusters to obtain the design of knowledge points that need to be improved.

Cluster analysis is one of the important techniques in the field of data mining, cluster analysis deals with data objects that are unknown, the process of cluster analysis is the process of grouping data objects in a given set into multiple clusters, the data objects within the same cluster have strong similarity, while the data objects between different clusters have dissimilarity, and the data objects in the set are divided into multiple clusters, this process is called cluster analysis, and cluster analysis is This process is called cluster analysis. Cluster analysis is widely used in the data processing of meteorological analysis, financial analysis, experimental analysis and so on.

K-means clustering algorithm is the most popular algorithm in cluster analysis, and many algorithms are derived from K-means based clustering algorithm such as standard K-means [28], Scalable K-means, EM, etc., with the aim of making the results of cluster analysis to be optimal. In K-means mean clustering algorithm, K is used as a parameter for the number of clusters, i.e., n data objects are divided into K clusters, so that the data objects within the clusters have a high degree of similarity and the data objects between the clusters have the lowest degree of similarity.

K-means mean clustering algorithm is defined as follows:

In the dataset B contains n data points i.e. B(x₁, x₂⋯x_n), k is the value of the number of expected clusters, M₁, M₂, M₃⋯M_K is a K subset of M and M_i ∩ M_i = Φ In K-means clustering algorithm, Euclidean distance is used to obtain the center of mass C_j closest to the x_i point.

Inputs to K-means mean clustering algorithm: 1)

Number of clusters K.

Algorithm function: K-means algorithm calculates the mean value based on the objects in the clusters.

Algorithm Output: K clusters are formed such that the mean value of the distance between the data objects in each cluster is minimized.

Steps of K-means mean clustering algorithm: 1)

Among the n data objects in the dataset, K random data objects are selected as the centers of mass for the K clusters.

The calculation of the distance is carried out by this clustering criterion function, where the K value is the number of classes, M_i is the representative of class C_i, and is the square of the Euclidean distance between the rest of the data objects in class C_i, O, and the representative of class C_i, M_i. 3)

Calculate the centers of gravity of each class and use these centers of gravity as the representatives of each class in the algorithm: (2)aj=∑i=1n1{C(i)=j}x(i)∑i=1n1{C(i)=j}x(i)$$aj = \frac{{\sum\limits_{i = 1}^n 1 \left\{ {{C^{(i)}} = j} \right\}{x^{(i)}}}}{{\sum\limits_{i = 1}^n 1 \left\{ {{C^{(i)}} = j} \right\}{x^{(i)}}}}$$

The K-means mean clustering algorithm is implemented as follows: 1)

Randomly select K different data objects from dataset B as representatives H, H₂, H₃⋯H_k of K classes C₁, C₂, C₃…C_k

||O−Hi||=min{||O−H||1≤j≤k}$$||O - {H_i}|| = \min \left\{ {||O - H||1 \leq j \leq k} \right\}$$// Finds the closest P_i to O and assigns O to the C_i to which H_i belongs.

For each class C_j(1 ≤ j ≤ k)

Hj=Mj=∑0∈CinX|Cj|$${H_j} = {M_j} = \frac{{\sum\limits_{0 \in {C_i}}^n X }}{{|{C_j}|}}$$// Select the center of gravity of the class as a representative of the class

E=∑i=1k∑0∈Ci||O−Mi||$$E = \sum\limits_i = {1^k}\sum\limits_{0 \in {C_i}} {||O - {M_i}||}$$// Calculate the clustering criterion function E. 3)

Until K class representatives no longer change.

The general clustering effectiveness is measured by the average quantization error q(c) as follows: (3)q(c)=1n∑i=1n(Xi,Cj)$$q(c) = \frac{1}{n}\sum\limits_{i = 1}^n {({X_i},{C_j})}$$

When the value of x_i ∈ D_j, q(c) is very small, the clustering of learning outcomes of students in electrical automation technology is better.

In this paper, we collected behavioral data from 1400 online students of operating system course, labeled their attributes by creating portraits, and performed correlation validation and cluster analysis for students’ experimental practice ability, and analyzed the dimensions containing the number of effective operations, online hours, after-class hours, and course hours. Some of the original samples are shown in Table 1. As can be seen from the table. Students with effective operation number above 500 also have longer after-class study hours.

First, four dimensions were extracted from the original sample to conduct W-test normal validation of single-dimension samples respectively, to determine whether the single-sample dataset conformed to the normal distribution. Then, the single-sample data conforming to the normal distribution and the composite scores were subjected to Pearson correlation coefficient validation to verify their correlation. Finally, the clustering algorithm was used for clusterability analysis. Among them, the normal validation and Pearson correlation coefficient validation are shown in Table 2.

The result values of W-test normal validation for the four dimensions are all significantly greater than 0.05, and the distribution of the sample data are all in line with the normal characteristics. Meanwhile, it can be seen that the Pearson correlation coefficients of the three dimensions of total study hours, after-school hours, and course hours of electrical automation technology and the comprehensive grade of the course are in the range of 0.8 to 1.0, among which the correlation coefficient of course hours of electrical automation technology is the largest, which is 0.8935, which proves that there is a very strong correlation between these three dimensions and the comprehensive grade of automation technology course. Moreover, the Pearson correlation coefficient between the number of effective operations in electrical automation technology and the comprehensive grade is 0.6396, which is between 0.6 and 0.8, proving that there is a strong correlation between the two.

Based on the sample data of the four dimensions mentioned above, which have some correlation with the comprehensive performance of the electrical automation technology program, the next step is to carry out the analysis of the K-Means clustering algorithm. The K-Means algorithm settings are compared horizontally to see the classification effect, and the analysis is carried out. In the process of performing iterations, the sum of error squares is used for the adjustment of the optimal number of iterations. Figure 1 shows a two-dimensional cut-away of the data amplification effect in the case of optimal iterations. At the same time, the clustering effect of 1200 students was combined with the single-point data of student behavior for cluster analysis, and it was found that each cluster implies the corresponding behavioral information. For example, students with better final grades generally present high number of effective operations and high after-school hours in electrical automation technology labs. While students with lower final grades will generally present the characteristics of low number of effective operations in the experiments of the electrical automation technology program and low online hours on the learning platform. Table 3 further demonstrates the number of the 4 categories, behavioral characteristics, etc. of the cluster analysis.

Figure 1.

Two-dimensional section of the k-means

The entire student collection was analyzed proportionally, and the results of the analysis are shown in Table 4, most of the students were distributed in Type 2 and Type 3, and a small portion of them were distributed in Type 1 and Type 4, and the distribution was relatively normal, from which it can be concluded that the experimental teaching of the electrical automation technology major should focus on strengthening the requirements of the students’ after-school experimental practice as a way to increase the number of the student population of Type 1.

At the same time, analysis from the perspective of teaching management reveals that the number of Type 1 students with good overall performance is 155, accounting for 12.92%. The number of Type 4s that performed poorly overall was 156, which is not a significant difference. Based on this, some type 1 and type 4 students can be arranged in the same experimental class or the same group team to form the teaching mode of “one with one, good with bad”, which can stimulate the learning interest of type 4 students, improve the overall practical ability of the students and improve the teaching effect.

Clustering algorithms can only classify known samples, rather than enabling prediction of unknown samples, so that the analysis of student learning behavior will always lag behind. Each time, the response can only be studied after the negative learning behaviors are produced, which is obviously not the effect that educators want to achieve.

Logistic regression model [29] as a kind of machine learning model, just can realize the above needs, after reasonable training, not only can classify the existing students’ satisfaction with the correctional reform, but also can make certain predictions on the unknown students’ satisfaction, moreover, its whole process is easier to realize automated operation, which greatly reduces the consumption of human and material resources. Combined with the process of teaching reform in electrical automation and the logistic regression model itself, the flow chart of the model designed is shown in Figure 2.

Figure 2.

Flowchart of the machine learning model

By constructing a logistic regression model, manually analyzed sample data is used as a training set to eliminate the more unusual samples, i.e., to prevent isolated points from affecting the learning system in a bad way. This training set is used to train the model, stabilize it and test it periodically with a test set of unknown samples. So that it can be applied to the prediction of unknown samples. In order to easily determine whether the results given by the model can be directly applied to the actual work of student satisfaction analysis, a credibility assessment system is added to the model, which will comprehensively assess the credibility of the final results based on the preset weights of each attribute. Credibility is a decimal number between 0 and 1.

There are many classifiers that can be used for classification in machine learning, and this paper applies logistic regression to try to explore student satisfaction classification.

Logistic regression models are usually binary models and are used for binary classification problems. However, it can be generalized to a multinomial logistic regression model for multi-categorization. This model is generally used to estimate the likelihood of something. Examples include the likelihood that a student will skip class, the likelihood that a doctor will own a cat, the likelihood that a driver will refuse to take a taxi, and the likelihood that a game will be downloaded by a user. The result of logistic regression is not a probability value in the mathematical definition. The result is usually used for weighted summation with other eigenvalues, rather than direct multiplication.

For a binary classification problem, using E_e(X) = 0 to represent the decision boundary, there is a probability that the sample label is of a certain type greater than 0.5 when E_e(X) > 0. When E_Θ(X) < 0, the probability that the sample label is of a certain type is less than 0.5, and the probability that the sample label is of a certain type is about 0.5 when E_e(X) is approximately equal to zero.

If one wants to know what the probability is that the sample label is of a certain type, this can be achieved by the sigmoid function, which can be applied to convert E_e(X) functions to probability functions in binary classification problems. From the curve of the sigmoid function on the coordinate scale, it can be seen that as x increases, the corresponding sigmoid value will be infinitely close to 1. And as x decreases, the sigmoid value will be infinitely close to 0.

In the dichotomous problem, the logistic regression function h_Θ(X) = sigmoid(E_Θ(X)) can be found, and the decision boundary can be viewed as the contour at h_e(X) = 0.5. The formula is shown below: (4)g(x)=11−e−x$$g(x) = \frac{1}{{1 - {e^{ - x}}}}$$ (5)hΘ(X)=g(EΘ(X))=g(ΘTX)=11+e−ΘTX$${h_\Theta }(X) = g({E_\Theta }(X)) = g({\Theta^T}X) = \frac{1}{{1 + {e^{ - {\Theta^T}X}}}}$$

Compared with linear regression, the variable y of logistic regression obeys the Bernoulli distribution, and the variable y of linear regression obeys the Gaussian distribution, so logistic regression has a lot of similarities with linear regression, and the logistic regression algorithm is a linear regression by removing the Sigmoid mapping function. It can be said that logistic regression is theoretically supported by linear regression, but logistic regression introduces nonlinear factors through the Sigmoid function, so it can easily handle 0/1 classification problems. So it is very suitable for automated classification of samples, especially for classification analysis of massive samples.

And for the case of multiple categorization, suppose the set of values of category Y is {1, 2, …, K} then the multinomial logistic regression model is: (6)P(y=k|x)=exp(wk⋅x)1+∑k=1K−1exp(wk⋅x)k=1,2,⋯,K−1$$P(y = k|x) = \frac{{\exp ({w_k} \cdot x)}}{{1 + \sum\limits_{k = 1}^{K - 1} {\exp } ({w_k} \cdot x)}}k = 1,2, \cdots ,K - 1$$ (7)P(y=K|x)=11+∑k=1K−1exp(wk⋅x)$$P(y = K|x) = \frac{1}{{1 + \sum\limits_{k = 1}^{K - 1} {\exp } ({w_k} \cdot x)}}$$

The likelihood function is: (8)∏i=1N∏k=1KP(yi=k|xi)yik$$\prod\limits_{i = 1}^N {\prod\limits_{k = 1}^K P } {({y_i} = k|{x_i})^{y_i^k}}$$

where P(y_i = k ∣ x_i) is the probability that the model will award sample x_i to category k when it is entered, and yik$$y_i^k$$ serves as an indicator function, 1 when k is equal to the labeled category of sample x_i, and 0 for the rest.

Taking the logarithm of the likelihood function and then taking the negative, we get L(w₁, w₂, ⋯, w_K−1) (abbreviated as L(w)), which is the final model parameter w₁, w₂, ⋯, w_K−1 to be trained such that the value of L(w) is minimized.

Because the dataset to be dealt with in this paper is a massive location sample set with a lot of attributes related to the determination, which cannot be dealt with by a simple binary classification model, the logistic regression model for the multiclassification case is used in this paper as a practicing model for machine learning in sample family clustering.

First of all, the preparation of data collection should be carried out, get the specified number of samples from the working sample set, and batch get the VM (virtual machine) and sandbox to extract the static and dynamic attributes with the relevant inverse tools, and the acquired data should be organized, and if it is the log class data, the corresponding attribute extraction should also be carried out. In this paper, a small program written in C# is used to extract attribute data from logs, and regular matching is used in the program.

Secondly, data preprocessing is performed to eliminate null data and extreme data, and quantize all attribute values into values represented by 0 and 1.

Finally, it is ensured that the sample sets used to train the logistic regression classifier have been manually categorized so as to ensure that the trained model has superior classification accuracy.

Once the above steps are completed, the classifier can be trained and then the accuracy of the results can be analyzed and the dataset can be modified and these steps can be repeated until the accuracy of the student satisfaction prediction results meets expectations.

This chapter explores and constructs a machine learning model based on logistic regression, which can not only realize the classification of massive student satisfaction samples required in the work of education reform, but also adjust the classification accuracy in real time based on the characteristics of machine learning and become more and more accurate according to the training of a reasonable training set, and moreover, it can realize the prediction of unknown samples. So the model obviously has better adaptability when facing situations such as the emergence of new samples.

Logistic regression, also known as logistic regression analysis, is a generalized linear regression analysis model commonly used in the fields of data mining, automatic diagnosis of diseases, and economic forecasting. Through logistic regression analysis, the weight of each independent variable can be obtained, so that we can roughly understand the factors with the greatest weight that lead to the occurrence of an event. At the same time, the likelihood of the occurrence of the event can be predicted based on the weight value of the factor.

In this paper, 200 students were selected as the dependent variable of whether they were satisfied with the overall feeling of the education reform of the electrical automation technology program in which they had participated, and the four public factors of practical training base selection (X1), internship process (X2), practical training process (X3), and intelligent manufacturing facilities (X4) derived from factor analysis were used as the independent variables. Logistic regression analysis was carried out using SPSS21 software so as to derive the magnitude of the influence of each variable on the students’ satisfaction with the teaching reform of the electrical automation technology program. After the completion of the analysis, the model needs to be tested to determine and analyze the fit of the model and the accuracy of the prediction, so as to test whether the model is reasonable and the accuracy of the model results, and the test results are shown in Table 5. The closer the value of Cox&Snell R² and the value of Nagelkerke R² is to 1, it means that the better the fit effect is, and the value of Cox&Snell R² is 0.304, and the value of Nagelkerke R² is 0.304, and the value of Nagelkerke R² is 0.304. Nagelkerke R² has a value of 0.409, which indicates that the whole model fit is average. It shows that the model estimates basically fit the data at an acceptable level, indicating that the model is applicable.

Table 6 shows the classification table of the logistic regression results, which is in the form of a matrix table to show the degree of matching between the predicted values of the model and the actual observations, and the degree of perfection of the model fitting can be evaluated by the correct rate of prediction, and the higher the correct rate means that the model is fitted better. From the results of the classification table, it is known that the correct rate of prediction reaches 83.2%, which indicates that the model is more satisfactory and acceptable.

In the results of the logistic regression analysis as shown in Table 7, it can be seen that the P-value of all the variables is less than 0.05, which passes the test at the level of significance, which means that the independent variables selected for the model have a significant effect on the dependent variable, so the variables can be used as the final variables of the model.

Based on the results in Table 7, the following binomial logistic regression equation is derived: (9)logist p=0.157+0.969x1+0.765x2+0.823x3+0.735x4$${\text{logist }}p = 0.157 + 0.969{x_1} + 0.765{x_2} + 0.823{x_3} + 0.735{x_4}$$

The base selection factor represents the influence of the training base environment on students’ satisfaction with the education reform of electrical automation technology, including five aspects: “transportation convenience of electrical automation technology training base”, “internal environment of training base”, “surrounding environment of training base”, “service attitude of training base personnel” and “service quality of training base personnel”. The model shows that the impact of this factor on student satisfaction is significant at the 95% confidence level, and the coefficient is 0.969, which is greater than 0, which has a positive impact, indicating that the higher the score of the base selection factor, the greater the students’ satisfaction with the education reform of electrical automation technology.

The internship process factor represents the students’ feelings in the internship process and the impact on the satisfaction of the education reform of the electrical automation technology major, including four aspects: “difficulty of internship content”, “interaction between teaching and students in the internship process”, “interaction between students” and “richness of internship content”. The model showed that the impact of this factor on student satisfaction was significant at the 95% confidence level, with a coefficient of 0.765, which was greater than 0, which had a positive impact, indicating that the higher the score of the internship process factor, the greater the students’ satisfaction with the education reform of electrical automation technology.

The training process factor represents the influence of training on students’ satisfaction with the education reform of electrical automation technology, including four aspects: “training schedule”, “proportion of training items”, “training quality” and “training practicality”, and the model shows that the impact of this factor on student satisfaction is significant at the confidence level of 95%, and the coefficient is 0.823, which is greater than 0, which has a positive impact, indicating that the higher the score of the training process factor, the greater the students’ satisfaction with the education reform of electrical automation technology.

The intelligent manufacturing facility factor represents the impact of intelligent manufacturing equipment and facilities on students’ satisfaction with the education reform of electrical automation technology, including four aspects: “intelligence of facilities”, “safety of facilities”, “hardware facilities” and “software facilities”. The model shows that the impact of this factor on student satisfaction is significant at the 95% confidence level, with a coefficient of 0.735, which is greater than 0, which has a positive impact, indicating that the higher the score of the intelligent manufacturing facility factor, the greater the students’ satisfaction with the education reform of electrical automation technology.

This paper validates and explores the field of teaching reform in electrical automation using machine learning techniques to effectively identify student learning behaviors, complete the clustering of student learning behaviors and their prediction of satisfaction with teaching reform.

The clustering analysis results of students’ learning behaviors show that most of the students are distributed in type 2 and type 3, i.e., the behavioral categories are average number of valid commands and long after-school hours, which can be improved by increasing the number of hours of after-school practical learning for students. The number of students with good performance is 12.92%, which is close to the number of students with poor performance. The two types of students can be arranged in study groups to form an excellent teaching mode with good performance to improve the learning and practical ability of the overall electrical automation students and improve the teaching effect.

Training base selection (0.969), internship process (0.765), practical training process (0.823), and intelligent manufacturing facilities (0.735) all positively affect students’ satisfaction with the teaching reform of electrical automation. Among them, the selection of practical training bases ranked first in the degree of influence of students’ satisfaction with the reform of electrical automation teaching, indicating that the better the development and construction of practical training bases, the higher the students’ satisfaction with the reform of electrical automation teaching. The degree of influence of practical training process, internship process and intelligent manufacturing facilities are the second, third and fourth respectively, suggesting that in the future reform of electrical automation teaching, efforts can be made on further improvement of intelligent manufacturing facilities to contribute to talent cultivation.