Marketing mix des services et fidélisation des clients dans les entreprises de téléphonie mobile : le cas de Mtn Cameroun

Abstract

Machine learning techniques are continuously taking more place in scientific and industrial domains, and that is also the case in mechanical engineering. This work deals with the setting up of a smart technique of finding the roughness of grinding pieces. A mathematical model based on the Artificial Neural Network (ANN) has been put in place. 32 experiments have been carried out, aiming at collecting the real value of the average arithmetic roughness (Ra) of the surfaces with the aid of the roughness tester SRT 6210, and the corresponding images of those surfaces with the microscope MV-900. Every surface was obtained with different feed rate, depth of cut and lubrication which were also used as input parameters in the neural network. The algorithm has been deployed in MATLAB R2017b. A neural system with 3 inputs, one hidden layer of size 10, the Levenberg-Marquardt backpropagation as activation function, the Mean square error as performance measurement and one output has been adopted. The algorithm was set using 28 experimental measurements where 50% were used for the training phase, 30% for the test phase and 20% for validation. The remaining 4 have been used to assess its behaviour with random entry data. The performance of the algorithm is satisfactory with an R² of 0.99892 for the training, 0.94252 for testing and 0.92305 for validation. The overall performance is expressed by an R² of 0.87221. The accuracy of the algorithm is estimated at 90,652%.