This study investigates the optimization of an integrated production–inventory system that consists of an original equipment manufacturer (OEM) supplier and an OEM brand company. The cognitive and skill learning effect, imperfect quality, and fuzzy random demand are incorporated into the integrated two-echelon supply chain model to minimize the total cost. We contribute to dividing the learning effect into cognitive learning and skill learning, we build a new learning curve to resemble the real complexity more closely and avoid the problem that production time tends towards zero after production is stable. In total, five production–inventory models are constructed. Furthermore, a solution procedure is designed to solve the model to obtain the optimal order quantities, and the optimal shipment size. Additionally, the symbolic distance method is used to deal with the inverse fuzzification. Then numerical analysis shows that the increase of the cognitive learning coefficient and skill learning coefficient will reduce the total cost of the production–inventory system. With the increase of the cognitive learning coefficient, the gap between the total cost of cognitive learning and skill learning, and that of Wright learning, correspondingly decreases consistently. However, with the increase of the skill learning coefficient, there is a consistent corresponding increase. The total cost of cognitive learning and skill learning shows hyperbolic characteristics. The important insights of this study for managers are that employees’ knowledge plays an important role in reducing costs in the early learning stage and humanistic management measures should be taken to reduce employees’ turnover. Compared with the skill learning training for production technicians, we should pay more attention to the training of cognitive learning.