Machine Learning (ML) is a group of computer-based techniques that can identify patterns within extensive collections of data. These templates can be used to learn various nonlinear process, mathematically model them, and predict outcomes. ANNs are one such tool. As name implies, an Artificial Neural Network (ANN) is network of computational processor known as nodes or artificial neurons. These are connected together to form a non-linear processing system. (Alpaydin, 2014)

A multilayer perceptron (MLP) is type of neural network that has multiple layers of neurons. The first layer, called the input layer, receives input data and processes it. The processed data is then sent to one or more hidden layers for further processing. Subsequent output layers perform final output calculations (Sarkar, 2009). The connection between the inputs and the outputs of a neural network is known as its architecture or topology (Martin T. Hagan, 1997). Configuration of an artificial neural network can be altered by modifying quantity of neurons present in hidden layers or by changing number of hidden layers altogether. For this particular study, an ANN configuration consisting of two input neurons and two output neurons was utilized.

Three sets of input data are typically divided by ANN: training, validation, and testing. During the training process, ANNs use the training set to learn patterns and develop their models. Validation set is used to fine-tune the model, ensuring that it can generalize well to new data. Once a model is fully trained, test set is used to evaluate its final performance and compare it to other models (R. Porrazzo, 2013).

In order to obtain a model which furnishes us with the output with least mean-squared error we have used a MATLAB code which provides us with a graph which have number of hidden layers on the x-axis and the value of Mean Squared Error (MSE) on the y-axis. The minima obtained from the graph is the value of the hidden layer which we are going to use to train our model. The corresponding graphs are:

## 4.1 Data Extraction

Plots, graphs, and diagrams are graphical representations used to describe raw data collected from observations in experiments and scientific publications. xyExtract software was used to extract raw data (from 2D plots) from published research articles. The extracted data is then used for ANN and simulation training to produce findings for reaching optimum concentration based on variations in output and input parameters at different operating circumstances, namely speed and time. The experimental data from (Sofie T. Morthensen, 2015)are used to train the ANN in this work.

## 4.2 Details of the simulations

The ANN model was built with MATLAB's Deep Learning Toolbox (Kim, 2017). A total of 176 experimental data findings are used for training, validation, and testing the Artificial Neural Network (Sofie T. Morthensen, 2015). Ultimate artificial neural network (ANN) model consists of three layers: an input layer having the same quantity of neurons as that of input parameters, a hidden layer with a variable number of neurons, and an output layer containing only one neuron, which is linked to a single output.

To determine the best topology for an artificial neural network (ANN), several models were trained with varying numbers of neurons in the hidden layer. Model that had the least mean squared error (MSE) was chosen to forecast output parameters. Using MATLAB, anticipated outcomes were graphed against input parameters, resulting in creation of three-dimensional surface plots.

The trained artificial neural network (ANN) is utilized in combination with MATLAB optimization toolbox to determine most efficient values of operating parameters. The ANN is executed in a separate MATLAB script from the optimizer. This script makes predictions of the concentrations for a particular set of input variables, which can be used to calculate objective function's value as output variables. The script is designed as a MATLAB function that is run in the MATLAB Optimization Toolbox to predict the optimal values for operating parameters. To maximize concentration value, simulations are carried out at different concentrations and operating times.