Project Goals
The goal of the project is to investigate methods for classifying data that monitor the behavior of a robotic system in order to locate data sets with signs of a fault.
The methods we will investigate are of the One-Class Classification type that allows to identify anomalies without their exact characterization.
By taking data measured from real robotic systems, we try to detect faults in the systems, and thus try to predict for future systems whether a fault is expected or not.
Chosen Solution
Firstly, the project involved the creation of a synthetic database. Two data sets were built – the first set consisted of sine signals at a 10kHz frequency, the second set was the sum of two sine signals at 10kHz and 1.2kHz. Noise was also added. Then, a FFT transformation of the signals was calculated.
We used two One-Class Classification algorithms: One-class SVM and K-Means.
We used the synthetic databases to learn and optimize the classification methods before using them for the real data.
The results were analyzed using binary classification, and by calculating the false negative rate and the false positive rate.
In the second stage, after improving the algorithms on synthetic data, real data from robotic systems was used. This database contained 8 different sets, each representing data taken from a robotic system, some with a fault and some without.
Set number 3 was compared with all the other sets. It is known that sets 3, 4, 5, 7 have similar characteristics, and sets 8-11 have similar characteristics.
Results
Synthetic database – We were able to bring both algorithms to produce good results for the synthetic databases, even with added noise, which is much larger than the amplitude of the sine signals.
Comparing both algorithms, the K-Means algorithm yields better results and has a much faster operating time than SVM.
Real database – The results in the databases from the robotic systems were inconclusive. In some cases, there was a good distinction between faulty databases and fault-free databases, and in other cases there was a more problematic distinction.