PSI - Issue 59

Victor Aulin et al. / Procedia Structural Integrity 59 (2024) 436–443

437

Victor Aulin et al. / Structural Integrity Procedia 00 (2019) 000 – 000

Keywords: Cylinder-piston group, Valvetrain, Slider-crank linkage, Engines

1. Introduction The need to exclude violations of the homogeneity of the sample is justified in the method of mathematical processing of the database of experimental data of the technical condition of automobile units and the most informative controlled parameters are selected (Hrynkiv et al., 2020; Aulin et al. 2018). The obtained empirical data were used to train models of the mathematical apparatus of artificial neural networks (ANNs). Assessment of their accuracy, at the stage of diagnosing of the technical condition, when recognizing defects in the resource-determining elements of car engines: cylinder-piston group, slider-crank linkage, valvetrain. The distribution by complex of technological operations of maintenance and repair is also used (Aulin et al., 2020). At the present time, the distribution of the complex of maintenance and repair operations is carried out according to the principle: each node, system and unit of machines after the diagnosing of the technical condition is accompanied by a diagnostic card, where the operator-diagnostician notes the detected defects. The next step is comparison of the diagnostic map with the technological map of each complex of maintenance and repair operations to select the appropriate route (Aulin et al., 2018). According to this approach, the human factor is the main reason for the appearance of defects of the 1st (false defect) and 2nd type (passing defect). The main condition for solving the task is that the number of complexes of maintenance and repair operations should cover all possible defects of resource-determining elements of the engine (Aulin et al., 2018; Viktorova 2012; Hevko et al., 2018). At approach of neural network classification, where the accuracy of defect recognition and their distribution the human factor has less influence. Basically, the selected structure, algorithms for building and training ANN are influential in the ANN method (Myhal 2018; Sierikov 2012; Tilniak et al., 2018; Order of the Cabinet of Ministers of Ukraine 1556-r, 2020; Aulin et al., 2020). At the same time, it is possible to achieve a reduction of intra production losses due to emerging errors at the stage of diagnosis and during the distribution of resources of the primary elements according to the complex of technological maintenance and repair operations . The determination of the size of the training sample for quality training is an important stage of experimental research using the method of the ANN mathematical apparatus. An insufficient number or contradiction of data in the training set will be reflected in a large number of errors in the classification of defects and distribution of the resource-determining elements of the car engine at the testing stage (Ostrovska et al., 2023; Patan et al., 2008; Dmytriv et al., 2023). At the present time, there is no generalized way to determine the required number of examples for high-quality training of ANN, therefore this task still does not have an unambiguous solution (Kukurudziak 2012; Sobchuk et al., 2021; Sobchuk et al., 2019; Molodan et. al., 2021; Maruschak et al., 2016). The hypothesis shows that for learning of an adequate model of the ANN mathematical apparatus, a sufficient number of training examples P is equal to the product of the number of neurons of the input and output layers: Nx – the number of neurons of the input layer; Ny – the number of neurons of the output layer. The aim of this work is to build a block diagram of the mathematical processing of diagnostic information of resource-determining elements (the cylinder-piston group, the slider-crank linkage, the valvetrain), determine the maximum errors during training and testing methods of ANN models, and build graph models of ANN for recognizing their defects. 2. Methodology At the stage of experimental research, the construction of ANN models was carried out in IBM SPSS Statistics software. This software product is chosen due to the fact that it is one of the leaders in the global market of analytical platforms, which implements most modern approaches to database analysis and its processing. YMZ-238 engines were the basis of research. Resource-determining elements (the cylinder-piston group, the , P Nx N y   (1)

Made with FlippingBook - Online Brochure Maker