PSI - Issue 59

Victor Aulin et al. / Procedia Structural Integrity 59 (2024) 444–451 Victor Aulin et al. / Structural Integrity Procedia 00 (2019) 000 – 000 7

450

Input layer of neurons l =0

The last hidden layer of neurons

The first hidden layer of neurons

Output layer of neurons

l =1

l = L -1

l = L

x 1

Y 1

p 1

p 1

p 1

x 2

Y 2

p 2

p 2

p 2

x 3

...

...

...

...

Y L

p l

p L -l

p L

x i

Fig.3. The diagram of a multilayer perceptron in the artificial neural network model.

To recognize and predict combinations of defects in components of units, systems, and assemblies of automobiles using the ANN method, an algorithm has been developed, the flowchart of which is presented in Figure 4.

START

Step1. Initial database collecting (DB1) based on combinations of parameters of units, systems, and assemblies of automobiles.

Step2. Initial data normalization within the range of values of the activation function Y = f ( S ), using the algorithm (fig.2)

Step3. The number of neurons in the hidden layer: l =2, … , L -1 and determining their upper maximum value R in the hidden layer

Yes

Step4. Has the number of neurons reached the upper maximum value R ?

No

Step5 . The ANN method is trained using the error reverse propagation methods

Step6. Assessment of defect detection accuracy according to the structural-causal model of the interrelation of parameters in units, systems, and assemblies of vehicles based on the test sampling

Yes

Step7. Has the test sampling been successfully recognized?

No

Step8. An increase in neurons number occurs

Step9. Each such combination is assigned a specific set of maintenance and repair operations. The obtained results are then used to create a separate database (DB2)

FINISH

Fig. 4 . The flowchart of the algorithm for applying the ANN method to solve the problem of recognizing and predicting defects in components of units, systems, and assemblies of automobiles.

To distribute units, systems, and assemblies of vehicles based on maintenance and repair complexes of operations, a database (DB2) formed as a result of the previous algorithm's work (Fig. 2) has been used. In this database, all identified and predicted combinations of defects in the components of units, systems, and assemblies of