ANFIS Wear Prediction for Inconel Coated Steel
Uploaded by
Suresh Kumar RANFIS Wear Prediction for Inconel Coated Steel
Uploaded by
Suresh Kumar RCommon questions
Powered by AIANFIS was selected for its unique capacity to handle nonlinear relationships between input and output variables through its integration of fuzzy logic and neural networks. This hybrid approach allows for precise control over membership and consequent functions, making it superior in adjusting to complex dynamics typical in wear scenarios. Its ability to be finetuned via a hybrid learning algorithm, proven through high correlation with experimental data, further reinforces its selection over purely neural or fuzzy systems which may lack this dual adaptability .
The ANFIS model integrates both artificial neural networks (ANN) and fuzzy logic (FL) to predict dry sliding wear in mild steel coated with Inconel 718. It uses a hybrid learning algorithm combining back-propagation with the least square method, allowing it to map inputs through membership functions to outputs. This mapping is based on fuzzy 'if-then' rules of the Takagi-Sugeno type. By adjusting premise parameters in the first layer and consequent parameters in the fourth, ANFIS can finely tune its output to match the training data, achieving a high correlation coefficient of 0.9924 with experimental results .
ANFIS adjusts to changing parameters through its hybrid learning algorithm that dynamically modifies premise parameters in the first layer and consequent parameters in the fourth layer. These modifications are critical as they refine the model's internal rules and membership functions, enabling it to learn from patterns in the training data. This flexibility allows ANFIS to accurately predict wear despite variations in inputs such as load, speed, and distance, reflecting real-world conditions and complexities in the wear processes .
The correlation coefficient of 0.9924 signifies the high accuracy and reliability of the ANFIS model in predicting wear on mild steel coated with Inconel 718. This near-perfect correlation indicates that the model's predictions are almost identical to experimental results, demonstrating its effectiveness in capturing the complex interactions among the input variables and the resulting output. Such a high correlation reflects the model's excellent learning and prediction capabilities, validating the use of ANFIS as a robust alternative to conventional modeling techniques .
The methodology using ANFIS differs from conventional modeling approaches by incorporating both neural networks and fuzzy logic, thus providing a more flexible and adaptive model. Conventional models might not dynamically adjust parameters based on evolving input-output data relationships, while ANFIS fine-tunes parameters through its hybrid learning algorithm. This ability to model nonlinear relationships and adjust rules based on fuzzy 'if-then' structures allows ANFIS to predict with a high degree of accuracy, as seen in its correlation of 0.9924 with experimental results .
Using a training data set is crucial as it allows the ANFIS model to learn from actual input-output patterns, adjusting its rules and membership functions for accurate predictions. This study utilized 76 data points for training, which informed the tuning of premise and consequent parameters, ultimately leading to a model that closely aligns with experimental results. The testing data set's correlation to experimental values underscores the training set's effectiveness in enhancing the model's predictive accuracy, which is demonstrated in the correlation coefficient of 0.9924 between predicted and actual wear .
Triangular membership functions in the ANFIS model define the fuzzy logic parameters for input variables, such as load, speed, and sliding distance, by assigning linguistic values in terms of grades varying from zero to one. A value of zero indicates non-membership, and one indicates complete membership in the fuzzy set. For this study, three membership functions were generated for each input to accurately model the influence of these variables on wear prediction .
Plasma spray technology enhances the wear resistance of mild steel by coating it with materials like Inconel 718, which are harder and more durable. This process involves melting coating powders and depositing them on the steel surface to form lamellae, providing superior hardness and reducing wear when the steel rubs against harder substances. In this study, plasma spray is critical as it sets the stage for the wear prediction, wherein the coated surface's interaction with operational variables is modeled using the ANFIS system to predict wear accurately .
Challenges in implementing ANFIS in wear prediction models include selecting appropriate membership functions, ensuring accurate training data, and tuning the hybrid learning algorithm for optimal results. Additionally, managing the computational complexity due to multiple inputs can be difficult. These challenges can be mitigated by meticulous data pre-processing, choosing the right mix of triangular membership functions, employing comprehensive datasets for training, and iterating with various parameters settings to fine-tune the model. Continuous validation against experimental data helps in maintaining accuracy .
The ANFIS rule generation process enhances predictive capability by formulating 27 specific fuzzy rules that relate inputs (like load, speed, and sliding distance) to the output (wear). The system utilizes a Sugeno type fuzzy inference system which models complex relationships through these rules, enabling precise alignment of predicted with experimental wear values. By using a training data set and modifying premise and consequent parameters, ANFIS efficiently reflects real conditions and interactions, offering excellent prediction accuracy .