New trends in the management of production costs are associated with the development of Machine learning, taking into account the human factor. They often use a bifurcation procedure that assigns costs to one of two categories. For this, training of pattern recognition is applied. Systems engineering allows for more sophisticated partition learning models that categorize costs into one of four categories (quartering). For this, two bifurcation training models are integrated. Such quartering is used by the management of the company to stimulate personnel responsible for costs. However, if the management is not aware of the minimum costs, undesirable activity of the personnel is possible. In this case, personnel, as an active element, chooses costs so as to maximize own incentives. For this reason, the real costs of the company may be higher than the minimum. Therefore, traditional machine supervised learning algorithms for pattern recognition can be ineffective. To solve this problem in the face of uncertainty, a mechanism has been proposed that includes a procedure for training of quartering with the help of instructions from a trainer, as well as a procedure for stimulating. Sufficient conditions for the synthesis of such a mechanism are found, under which the costs are minimal. The use of this mechanism is illustrated with the example of training of quartering and incentives to reduce the cost of re-equipment of locomotives.