PSI - Issue 59

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

438

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

slider-crank linkage, the valvetrain) were subject to diagnosis and in the process indicators characterizing their technical condition were recorded (Molodan et al., 2021; Qing et al., 2020). The obtained results were used as a test data set to check the adequacy of the built ANN model and the accuracy of training. The received measurement results were subjected to increased requirements regarding their reliability. In order to ensure the necessary level of reliability, gross errors (clogging of data), primarily associated with a violation of the homogeneity of the sample of the obtained measurement results, were excluded. For this, we used the Titien-Mur criterion ( E k -criterion), which is a generalized consequence of the Grubbs criterion, in order to prevent the presence of two or more rare adjacent data outliers. The maximum value of the criterion sample was determined by the formula:

2

   n k i 1

  

  

max

 x i x k

,

max

(2)

E k



  n

2





 x i x

1

i

n k i     1 max – is the arithmetic mean of observations after filtering k maximum values; x – the x i

where

x k

n k 

arithmetic mean value of the entire sample. To remove outliers from the minimum values, the selection of the criterion takes the form:

2

   n k i 1

  

  

min

 x i x k

min

,

(3)

E k



  n

2





 x i x

1

i

n i k

x i

x k     1 min – is the arithmetic mean of observations after filtering k minimum values.

where

n k 

Building structural-consequential models of the main resource elements (the cylinder-piston group, the slider crank linkage, the valvetrain) basing on the results of research was found that the controlled parameters have significant heterogeneity in the requirements for reliability and unequivocal determination of the technical condition of the control object. In order to form a rational set of control and diagnostic parameters, it was decided to reduce the obtained space by eliminating non-essential controlled parameters by evaluating the degree of correlation between them based on the received empirical data. The block diagram of the research methodology in this work is presented in fig. 1. Each controlled parameter is represented by a separate neuron in the input layer of the ANN, and each possible defect of a certain engine element is a separate neuron in the output layer of the ANN. The set of signals of neurons of the output layer will form a certain combination of defects.