The design and research of safety in vehicle collision involves the characterization of mechanical properties of several materials with large deformation. The aim is to provide the method and the precise data of mechanical properties of materials in the simulation of car structure and the collision of passengers, such that the design depends more on virtual methods. While in real engineering applications, the characterization experiment of the materials cannot satisfy the requirements of the inputs of the constitutive model of the materials due to the complexity of the material environment. The constitutive model cannot reflect the complex environment of the material accurately as well. For example, the property of the material in the thermal effect area around the solder joint is complex. The deformation and destruction of active particles, the platinum and their adhesive interface are hard to characterize. In the tensile experiment of Knee ligament with large deformation, the geometry pattern or the geometry transformation in the loading process of the sample cannot be obtain precisely. etc. Generally, we can find lots of experiment data, nearly complete with some hybrids, but it does not meet the requirement of the fine response simulation. For example, we cannot precisely predict the position and time of the occurrence of the structure crack during the car collision. We cannot predict when a car collide with the pedestrian, which will unload first, the Knee ligament or the shinbone? In this lecture, we will discussed how to use big data and machine-learning to find needed information from large amounts of non-ideal experiment data.