Clarifying Data Analitics Concepts

Clarifying Data Analitics Concepts

1. Manufacturing now has the opportunity to gather large quantities of data, coming from different areas, such as product and process design, assembly, material planning, quality control, scheduling, maintenance, fault detection and cover all the product life cycle phases.

   • There´s a need for analytical information system is growing, in order to explore datasets and discover useful and often hidden information
     • Given the importance of extracting value from data, i.e. to send the right information to the right person in the right moment, which is boosted by the digitization trends that are permeating the manufacturing industry context, it is clear that a deeper knowledge of the data analytics tools and concepts is highly desired also in the industrial engineering field.

2. BACKGROUND ON MAIN CONCEPTS

   • The field of statistics is always in evolution, due to the arising problems that science and industry are bringing to the general attention. With the advent of computers, statistical problems exploded both in size and complexity and some new areas of research were developed. Nowadays it is common to hear about data mining, machine learning and data analytics, even if the real meaning of these emerging terms is not clear enough.

3. DATA MINING

   • The extraction of implicit, previously unknown and potentially useful information from data. The idea is to construct numerical algorithms that investigate datasets to extract hidden patterns and regularities. Patterns that reveal strong structures can be used to predict future outcomes. In data mining, the data is stored electronically and the search is automated through computers

4. MACHINE LEARNING

   • The field that developed most of the algorithms to find and describe structural patterns in data

5. DATA ANALYTICS

   • Or equivalently data analysis, is a trend word, since it is strictly connected with the buzzword “Big Data”. Actually we can say that data analytics is a more comprehensive term respect to data mining, since incorporates a great deal of statistical thinking. In fact, as showed by the doted oval in Fig. 1, data analytics is composed both of data mining (and its technical algorithmic counterpart in machine learning) and statistical models to clean data at the beginning and to validate rules at the end. Data analytics is therefore composed of: initial standard statistical techniques, such as visualization of data, selection of attributes, discarding outliers etc. in order to construct the initial example set; then, machine learning algorithms construct rules and statistical tests are used to validate them

6. AI

   • AI has been developed in the field of computer science, and deals with computer programs that possess own decision making capability to solve a problem of interest. AI systems can perform actions such as perception, interpretation, reasoning, learning, communication and decision making to arrive to a solution for a given problem, imitating the intelligent behavior of human expertise.

7. BIG DATA

   • Big Data refer to a type of datasets, in which variables are different in type and nature, for this reason, they compel to be analyzed with a number of specific techniques, models and algorithms, that belong to the Data Analytics, Data Mining and Machine Learning, in this way being directly linked with the main concepts of this paper.

8. EXTRACTION OF VALUE FROM DATA

   • The extraction of value from data is a new challenge that companies are now experiencing. Consequently, the need for analytical information systems is ever-growing to guide corporate decision-making, during the different phases of the product process, such as production, maintenance, logistic and quality control. Manufacturing companies need to embrace data analytics to remain competitive in the global market. By incorporating predictive analytics, it is possible to address unobservable problems, such as machine degradation and hidden defects when used in conjunction with manufacturing system data.

9. MACHINE LEARNING TECHNIQUES IN SMART MANUFACTURING

   • The diversity of data mining tools provides great opportunities, but the profusion of different options may cause some confusion. In the following we present a list of the machine learning techniques that are currently used in manufacturing.
     • Supervised Learning
       • Linear Methods for Regression: A linear regression model assumes that the regression function is linear in the variable inputs X1, … Xp. Multivariate linear regression is a generalization of linear regression, by considering more than one dependent variable
         • Linear Method for Classification: The linear classifier uses the variable’s characteristics to identify which class (or group) the variable belongs to.
           • Decision Tree Based Method: A “divide-and-conquer” approach to the problem of learning from a set of independent instances leads naturally to a style of representation called a decision tree.
             • Neural Networks: The central idea is to extract linear combinations of the inputs as new features, and then model the target output as a nonlinear function of these features. By a trial and error method, the network learns the correlations between the input and output pairs and then applies them to predict unknown output for a new set of data.
               • Unsupervised Learning
                 • Principal Component Analysis: PCA converts a set of observations of possibly correlated variables into a set of values of linearly uncorrelated variable.
                   • Association Rules: The goal is to find joint values of the input variables that appear most frequently in the dataset, i.e. to predict combinations of attributes.
                     • Cluster Analysis: Cluster analysis implements the grouping or segmentation of different objects into subsets or clusters as primary goal, meaning that objects in the same cluster are more similar than other objects assigned to other clusters.
                       • Self-Organizing Maps (SOM): A SOM map consists of components called nodes or neurons. Each node is associated with a weight vector of the same dimension as the input data vectors and a position in the map space. The nature of SOM makes it possible to be utilized for various industrial application, such as dimensionality reduction, data visualization, process monitoring