IEEE Consumer Electronics Magazine - July/August 2022 - 89

Figure 4. Comparative study.
Figure 3. Confusion matrix for AIRS 2. A: GOOD. B:
Air in fluid. C: Brake oil spill. D: Even inner brake pad
wear. E: Even brake pad wear inner and outer. F:
Uneven inner brake pad wear. G: Uneven brake pad
wear inner and outer. H: Drum brake mechanical
fade. I: Drum brake pad wear. J: Reservoir leak.
In the misperception matrix, elements in the
first row represent the number of datasets that
belong to the particular class. The first column
(except the diagonal element) represents the
miscategorized datasets. In the first row (row A
from left to right), out of 55 GOOD condition
datasets, all data were correctly categorized as
GOOD condition. Hence, the first (diagonal) element
is 55 and all other elements are marked as
zero, which means there is no miscategorization
among them. In the eighth row, the diagonal
component represents the data that are correctly
categorized. Out of 55 considered data, 52
datasets were correctly categorized as inner
brake pad even wear - inner (DPWI). The nondiagonal
component in the eighth row represents
the dataset that belongs to the other fault
condition. One data is misclassified as GOOD,
one data is miscategorized as DPWI, and one
more data is misclassified as DRMF. In this fashion,
the classification accuracies were found and
compared.9 Overall, AIRS 2 produced a maximum
classification accuracy of 97.82% (538 data
out of 550 datasets). AIRS uses ten-fold cross-validation
for finding the accuracy.
DETAILED ACCURACY BY CLASS
The summary of the categorization accuracy
of various versions of AIRS algorithms shows the
July/August 2022
AIRS 2 gives maximum classification accuracy
when compared with AIRS 1 and AIRS 2 parallel
versions. Even though the results of all three
algorithms are quite related in terms of classification
accuracy, AIRS 2 algorithm is little ahead
in terms of performance with better generalization
ability for improving the accuracy. The classification
accuracy of the AIRS 2 is validated
using the detailed accuracy by class as shown in
Table 2.
The individual class accuracy is identified
using true positive rate (TPR), precision, and
false positive rate (FPR).9 The TPR must be one
for an ideal classifier and the FPR should be
zero. Referring to Table 2, TPR is almost closer
to one and FPR is almost closer to zero. Precision
and recall are also very closer to one.
Hence, the considered AIRS 2 algorithm is suitable
for making the OBD model.
COMPARATIVE STUDY
The accuracy of AIRS family algorithms is
compared with the bench-marked algorithms,
such as CSCA, JRIP, fuzzy, and DT.9 Figure 4
shows the comparative study. It shows that AIRS
2 is superior to others. AIRS is a self-organizing
model that does not require any effort to optimize
the system parameters.
OBD TOOL DEVELOPMENT
Based on the categorization accuracy, we
develop an onboard diagnostic tool (OBD) to
indicate the condition of the hydraulic brake system
under dynamic conditions.
A set of rules suggested by AIRS 2 was used
for developing the LabVIEW-based graphical
user interface. The LabVIEW program is coded
89
IEEE Consumer Electronics Magazine - July/August 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - July/August 2022
