Fig. 1.
Analog circuit fault diagnosis procedure.
Fig. 2.
Test node selection.
Fig. 3.
SVM optimization.
Fig. 4.
Flowchart of exhaustive search.
Fig. 5.
Sallen key BPF.
Fig. 6.
Frequency response characteristics of BPF.
Fig. 7.
BPF individual test node ranking.
Fig. 8.
BPF accuracy of classification single test node.
Table 1.
SVM hyper parameters for BPF - single node testing.
| Kernel | Accuracy [%] | ||
|---|---|---|---|
| Random search | linear | 65 | C = 100 |
| rbf | 80 | C = 100, γ = 0.012 | |
| poly | 55 | C = 100, d = 2 | |
| sigmoid | 80 | C = 100, γ = 0.012 | |
| Bayesian search (number of iterations = 100) | linear | 70 | C = 223.64 |
| rbf | 80 | C = 63.65, γ = 1.82 | |
| poly | 75 | C = 6.01, d = 3 | |
| sigmoid | 75 | C = 338.98, γ = 0.575 | |
| PSO (number of iterations = 100, particle size = 10) | linear | 65 | C = 412 |
| rbf | 80 | C = 921.4, γ = 6.9 | |
| poly | 55 | C = 844.5, d = 2 | |
| sigmoid | 70 | C = 307.3, γ = 0.12 | |
| Exhaustive search | linear | 65 | C = 100 |
| rbf | 90 | C = 100, γ = 0.0127 | |
| poly | 70 | C = 1000, d = 2 | |
| sigmoid | 55 | C = 1000, γ = 0.1 |
Fig. 9.
BPF test node ranking.
Fig. 10.
BPF accuracy of classification - two test nodes.
Table 2.
SVM hyper parameters for BPF.
| Kernel | Accuracy [%] | ||
|---|---|---|---|
| Random search | linear | 95 | C = 100 |
| rbf | 95 | C = 100, γ = 0.0085 | |
| poly | 65 | C = 100, d = 3 | |
| sigmoid | 80 | C = 100, γ = 0.0085 | |
| Bayesian search (number of iterations = 100) | linear | 92 | C = 1000 |
| rbf | 95 | C = 1.913, γ = 0.9 | |
| poly | 90 | C = 86.32, d = 3 | |
| sigmoid | 90 | C = 1000, γ = 0.0417 | |
| PSO (number of iterations = 100, particle size = 10) | linear | 96 | C = 831 |
| rbf | 95 | C = 200.42, γ = 6.2 | |
| poly | 95 | C = 256, d = 3 | |
| sigmoid | 85 | C = 218.57, γ = 0.111 | |
| Exhaustive search | linear | 100 | C = 1000 |
| rbf | 100 | C = 1000, γ = 0.00857 | |
| poly | 95 | C = 1000, d = 2 | |
| sigmoid | 45 | C = 1000, γ = 0.1 |
Fig. 11.
State variable filter.
Fig. 12.
Response of the LPF of the SVF.
Fig. 13.
Response of the HPF.
Fig. 14.
Response of the BPF.
Fig. 15.
SVF single test node ranking.
Fig. 16.
SVF three test nodes ranking.
Fig. 17.
SVF SVM classification accuracy.
Table 3.
SVF SVM hyper parameters - single node testing
| Kernel | Accuracy [%] | ||
|---|---|---|---|
| Random search | linear | 22.22 | C = 4.4 |
| rbf | 11.11 | C = 100, γ = 0.01 | |
| poly | 11.11 | C = 35.93, d = 2 | |
| sigmoid | 22.22 | C = 100, γ = 0.01 | |
| Bayesian search (number of iterations = 100) | linear | 33.33 | C = 1.179 |
| rbf | 16.67 | C = 3.08, γ = 0.243 | |
| poly | 5.56 | C = 74.63, d = 2 | |
| sigmoid | 16.67 | C = 1000, γ = 0.0013 | |
| PSO (number of iterations = 100, particle size = 10) | linear | 22.22 | C = 887.42 |
| rbf | 22.22 | C = 267, γ = 0.101 | |
| poly | 16.67 | C = 256, d = 2 | |
| sigmoid | 16.67 | C = 154.4, γ = 0.022 | |
| Exhaustive search | linear | 22.22 | C = 100 |
| rbf | 50 | C = 100, γ = 0.092 | |
| poly | 50 | C = 1000, d = 2 | |
| sigmoid | 22.22 | C = 1000, γ = 0.01 |
Table 4.
SVF SVM hyper parameters.
| Kernel | Accuracy [%] | ||
|---|---|---|---|
| Random search | linear | 22.22 | C = 12.9 |
| rbf | 50 | C = 100, γ = 0.034 | |
| poly | 27.78 | C = 100, d = 2 | |
| sigmoid | 27.78 | C = 100, γ = 0.01 | |
| Bayesian search (number of iterations = 100) | linear | 55.6 | C = 157.8 |
| rbf | 66.67 | C = 1000, γ = 0.207 | |
| poly | 66.67 | C = 1000, d = 3 | |
| sigmoid | 44.44 | C = 769.83, γ = 0.0086 | |
| PSO (number of iterations = 100, particle size = 10) | linear | 50 | C = 191.37 |
| rbf | 50 | C = 267.13, γ = 0.172 | |
| poly | 37 | C = 152, d = 3 | |
| sigmoid | 27.78 | C = 261.3, γ = 0.0158 | |
| Exhaustive search | linear | 88 | C = 100 |
| rbf | 90 | C = 1000, γ = 0.046 | |
| poly | 89 | C = 1000, d = 3 | |
| sigmoid | 28 | C = 1000, γ = 0.01 |