Análisis estadístico del desgaste de la herramienta de mecanizado en la planta de motores de Ford

Abstract

[ES] En sectores como el de la automoción, generar un producto final de alta calidad y a la primera es fundamental. Para ello, es necesario aplicar técnicas estadísticas y técnicas de monitorización online para la detección precoz de fallos que ayuden a obtener una producción estable y libre de defectos. En la planta de motores Ford se han instalado unos sensores en una máquina de mecanizado capaces de capturar datos en continuo sobre las variables del proceso.

En primer lugar, sobre estos datos recopilados se han aplicado diversas técnicas estadísticas como PCA, modelos clustering, Random Forest y CART y los resultados obtenidos concuerdan y muestran que la variable torque es la variable más correlacionada con el desgaste de la herramienta de mecanizado. Sobre la variable torque se le ha aplicado el modelo Lee-Carter que nos ha permitido entender el comportamiento del torque a lo largo del tiempo.

En segundo lugar, se han monitorizando las variables del proceso mediante los gráficos de control de calidad multivariantes T2 y SPE. También se han aplicado los gráficos MCUSUM y MEWMA para monitorizar el torque de los distintos ejes.

Todas estas herramientas estadísticas nos han permitido pasar del mantenimiento preventivo que se realiza actualmente en la factoría sobre las herramientas de mecanizado a un mantenimiento predictivo.

[EN] The Ford engine plant is machined the different parts of an engine and then assembled. The engine plant is divided into different production lines. The name of each line corresponds to the piece that is machined: cylinder heads, crankshafts, camshaft, blocks, connecting rods ... This project is going to be carried out on the line responsible for machining the block of an engine. Within this line is the machine called operation 110 and is responsible among other functions of boring the 4 cylinders of the engine. The boring is the machining operation that is performed in holes of pieces already made to obtain greater dimensional accuracy. The cylinder is the part of the engine where the piston moves. Therefore it is vitally important for the operation of a car that these cylinders are very accurate. To avoid having parts with incorrect diameters to the tool that performs these machining has been estimated a very short life time. Every 500 cycles the operators must stop the machine and change the tool. This causes production to be stopped many times without really being necessary. In addition, another problem is that we are discarding many tools are changed without these have really reached the end of its useful life. Therefore, we want to optimize the useful life of the machining tools. In this way we would reduce production costs. To this end, sensors have been installed capable of measuring different parameters. Also, in the last station, operation 110 measures the diameters it has made. These diameters are stored in a database of a computer installed in the plant. It would be vital to collect this data for our database. In this project we will use machine learning techniques such as clustering, logistic regression, neural networks, decision trees, Naive Bayes, DOE ...