Possibilities of using the data of electronic systems of agricultural machinery for building predictive models

Improving the reliability and non-failure operation of equipment is the most important task of the maintenance system. More and more modern agricultural machines are equipped with remote control systems and sensors, which opens up new opportunities for predicting the failure of mechanisms. However, a variety of data coming from the intelligent systems of modern machines may have heterogeneous formats, be unstructured or partially structured, which does not allow to unambiguously determine their usefulness for building models and predicting malfunctions. In this article, using the example of data coming from sensors of Rostselmash combine harvesters, the hypothesis about the possibility of their processing for the machine learning purposes and the construction of predictive models in the future was tested. To study the data, the main methods of predictive analytics were used - methods of mathematical statistics, modeling: a descriptive statistical analysis was carried out, the main statistical estimates were calculated, regression and correlation analysis of available data was performed. Based on the results of regression analysis, taking into account significant parameters, regression models were constructed that allow predicting the value of one variable with known values of dependent variables. The degrees of dependence of the indicators in the studied data were determined, according to the results of the correlation analysis, a high closeness of the relationship (the correlation coefficient is higher than 0.7) between the rotation frequencies of the shafts and augers of the aggregates and nodes was noted. Regularities in the changes in indicators in previous periods were revealed: when plotting the distribution of rotation frequencies of the shafts and augers of aggregates over time, according to the available data, synchronicity of the distribution of values in a conditionally normal operating mode was observed. The future results are predicted on the basis of the revealed patterns: the deviation trend of one of the symmetry coefficients distribution graphs of the rotational frequencies of the shafts and screws of the units over time from the normal position will signal a change in the characteristics of a particular working unit. The obtained results confirmed the prospects of predictive analytics of agricultural machines based on data from on-board systems.