Key PdM – Advanced Analytics Methods in the Mining Industry
The literature indicates that Random Forest (RF) -33%, followed by Artificial Neural Network (ANN), Deep Learning (DL) -27%, support vector machines (SVM), - 25%, k-means - 13%, and others - 2%, are the most commonly used ML algorithms in PdM method .
RF Algorithm in PdM
RFs or random decision-making forests are a group learning method for classifying, regressing, and other tasks that operate by constructing a multitude of decision-making trees at training time and generating a class that is the class model (classification) or the mean tree prediction (regression). In 1995, Tin Kam Flo from IBM launched RFs. As the name indicates, the RF produces a "forest" (ensemble) with multiple randomized decision trees and adds a simple average of their predictions. RFs showed good performance when the number of variables exceeded the number of samples (observations). RF is a supervised learning algorithm that is used for tasks of classification and regression.
ANN in PdM
ANNs or connectionist systems are loosely based computation structures that make up animal brains in biological neural networks. Such systems "learn" tasks by taking examples into account, usually without having to program task-specific rules. ANNs are smart, biological neuron-inspired computational techniques. An ANN consists of several processing units (nodes) that work quite simply. In general, such units are connected by the associated weight communication channels and operate with only the local data that is indicated by their connections. The intelligent behavior of ANNs is derived from the interactions between the network processing units. In many industrial applications, including soft sensing and predictive control, ANNs have been proposed and are one of the most common and applied ML algorithms. The main advantages of ANNs are the lack of expert knowledge to make decisions since they are only based on past data; even if the data are inaccurate, they are not compromised and can be used in real time without modifying their design at every update by creating accurate ANNs for a specific app.