Predicting parison shape for plastic fuel tank by using Gaussian process regression

• Gaussian process (GP) was used to predict parison shape in multilayer polymer extrusion for automotive fuel tanks. • A GP model was trained on parison shape data from constant die-gap extrusion tests. • GP predicted parison shape under variable die-gap, with 4.8% diameter, 17.0% thickness, 5.3% length errors. • Each prediction took 20 s, fast enough for on-site selecting process conditions and reducing trial-and-error. In the production of plastic fuel tanks for automobiles, predicting the viscoelastic deformation during the extrusion process of a multilayer system of different materials is a challenging task both theoretically and by numerical simulation. A Gaussian process regression (GPR) is applied to predict the parison shape in the fuel tank molding. First, a series of parison extrusion experiment is conducted at constant die-gap conditions to obtain the parison shape data, and the relationship with the extrusion conditions is estimated using GPR. The obtained GPR model is then used to predict the parison shapes under variable die-gap opening conditions as performed in blow molding of fuel tanks. We found that the parison diameter, thickness distribution, and total length could be predicted with an accuracy of 4.8%, 17.0%, and 5.3%, respectively. This thickness prediction error is less than the acceptable tolerance of 20%. The prediction time using the GPR model was approximately 20 s, which is shorter than the time required to extrude a single parison, making it suitable for determining and reselecting extrusion parameters during the operation on the production site.