Inverse optimization of bath parameters providing the best coating properties in electroless nickel-based alloy coatings using artificial intelligence methods

Electroless nickel coating is a coating method that involves the electrochemical deposition of nickel onto a surface. In electroless nickel coatings, the formation of a coating layer on a surface is achieved through an electrochemical process. In this process, metal ions are attracted to a cathode surface using an electric current, and nickel is deposited there as a coating layer. This coating technology offers various advantages compared to traditional electrolysis methods. In electroless nickel coatings, achieving a more uniform and evenly distributed coating on the surface allows for a smoother appearance compared to other coating methods. Additionally, this coating method enables precise control over the coating thickness, enhancing conformity to desired specifications and value ranges. When compared to traditional electrolysis methods, electroless nickel coatings can be more effective, especially in bath coatings where surfaces have more complex and intricate physical characteristics. It enhances resistance to corrosion on surfaces and provides protection against wear and scratches by simultaneously increasing durability. Electroless nickel coatings stand out as a preferred method for surface protection, aesthetics, and durability in industrial applications. This method can be successfully applied to various components used across a wide range of industries. For instance, it is particularly favored in automotive parts due to its benefits in hardness, corrosion resistance, and aesthetics. Similarly, it is employed in electronic components, especially connectors and circuit boards, providing electrical conductivity and corrosion resistance on the parts. Another application area is in parts used in industrial machinery, where electroless nickel coating is preferred for its resistance to wear and longevity. In electroless coatings, bath parameters refer to input variables used to achieve desired properties such as hardness, thickness, corrosion resistance, etc. Depending on the type of coating to be applied, these parameters include variables such as temperature, duration, reducing agents, types, and quantities of powders that facilitate activation. Various methods can be employed for parameter optimization, and artificial intelligence techniques, particularly machine learning methods with high accuracy rates, have become widely adopted. In this study, a series of coating experiments were conducted for data acquisition, and the results obtained were compared with real-world outcomes. To determine the number of experiments, which parameters to use, and the levels of each parameter in the experiment, the Taguchi experimental design method was applied. Artificial Neural Networks were utilized to transfer the obtained experimental results into a computerized environment and facilitate the creation of new data in the future. Artificial Neural Networks (ANN) is a machine learning method inspired by the human nervous system. It has a structure consisting of input, output, and intermediate (hidden) layers. The primary goal is to minimize the error between its outcomes and real experimental results. Artificial Neural Networks in tasks involving pattern recognition, classification, and regression. These networks learn from data, adjusting weights and biases to improve accuracy over time. Their applications span diverse fields such as image and speech recognition, financial forecasting, and optimization problems. Artificial Neural Networks play a crucial role in modern artificial intelligence, contributing to advancements in various industries. As the aim of the study is to determine the bath type and parameters based on the targeted coating properties, the Genetic Algorithms method was employed in the second stage. Genetic Algorithms (GA) are optimization algorithms inspired by the process of natural selection and genetics. They use the mechanisms of evolution to find optimal solutions to complex problems. GAs involve generating a population of potential solutions represented as chromosomes, applying selection, crossover, and mutation operators to create new generations, and evaluating their fitness. This iterative process continues until a satisfactory solution is reached. Genetic Algorithms are widely applied in diverse fields such as optimization, machine learning, and design problems, providing an effective approach to finding solutions in complex and dynamic environments. This technique mimics the chromosome structure to reach a goal from a result, utilizing crossover rates and mutation features for optimization. Upon examining the values obtained in the study, it was observed that the error rates between the results of the developed model and the laboratory experiment results, especially calculated using methods like MAD, MAPE, MSE, RMSE, etc., were within acceptable limits. For instance, it was concluded that the MAPE value was below 10%, indicating a satisfactory performance of the developed model. The first aim of this study is to develop a model that predicts output values (wear, hardness and corrosion resistance, etc.) when input parameters (coating bath parameters) are given using artificial intelligence/machine learning techniques using coating data. In the next step, the developed model, which constitutes the main originality of the thesis, is obtained by using this model. In the developed model, a system that can suggest the bath type and parameters required to obtain a coating with the desired output values is aimed. Although studies have been conducted with artificial intelligence methods in the field of coating in the literature, this study differs from the existing studies in terms of the coating types considered and the artificial intelligence system to be applied. Currently, existing studies are artificial intelligence applications aimed at estimating properties using bath parameters, and this study is aimed at determining suitable baths and parameters of these baths that can provide the desired coating properties. The study aims to use these superior aspects of Artificial Intelligence methods in the coating field. In the coating field, a wide variety of bath types and different parameters are used to coat a wide variety of materials. Here, the hardness and wear resistance of the material change. If we consider these features as outputs, it is a matter of researching and choosing bath types and parameters to provide the desired output features. It is also possible to evaluate bath parameters as inputs. Therefore, it will be very useful to teach the computer the relationship between inputs and outputs in the coating field and to predict coating features by estimating the results of experiments that have not been done before without experimenting. Moreover, it is an innovative perspective for a system to suggest what kind of bath should be preferred and what the parameters of this bath should be to obtain a coating with the desired hardness and wear resistance in academia and industry. The main purpose of the thesis is to develop such a system, which is not found in the literature, using Artificial Intelligence techniques. While creating ANN models, a model with a multi-layer feedforward network is being studied. In the first stage, two-layer feedforward ANN models, one hidden layer and one output layer, are created. Later, different models will be tried to be designed by making changes to the parameters of these models such as the number of input layers, hidden layers and output layers, transfer and activation functions. These models were developed by using different training algorithms (Levenberg-Marquardt, BFGS Quasi-Newton, Resilient Backpropagation, Scaled Conjugate Gradient, Conjugate Gradient with Powell/Beale Restarts, Fletcher-Powell Conjugate Gradient, Polak-Ribiére Conjugate Gradient, One Step Secant, Variable Learning Rate Backpropagation etc.). In the model we developed, the GA method was integrated into the ANN model. In this section, the ideal bath parameters were obtained by specifying the limits of the bath parameters, the targeted purpose (minimization or maximization) for each output value (hardness, coating thickness, corrosion rate) and the weight (importance) of these output values. In all experiments, the weight of the output values was taken equally. While the maximization target was determined for the hardness and coating thickness from the output values, the minimization target was determined for the corrosion rate.