In previous articles we have seen how the algorithm is trained with thousands of brake noise events coming from real-life brake durability, so brake noise can be detected with a very high level of confidence. Currently, brake squeal is the noise being identified during this first phase of the project and is identified with a proper level of confidence, including frequency and SPL. In the second phase of the project, the algorithm will also be evolved to associate a rating to the squeal noise event detected. The algorithm will be capable to predict the subjective rating provided by a professional driver during standard driving or during specific noise research maneuvers. The real-time detection is currently under investigation and could affect the resolution of the spectrogram to be used to train the algorithm and to detect the brake noise. However, the current level of the study does not currently show any predictable problem that could arise when the machine learning algorithm is embedded within a real-time system. Other brake noises should also be identified, even if less amount of data is available when compared with brake squeal. The study shows an alternative method for automatic noise detection and shows the possibility of automatically rating the brake noise. Real-time detection is also investigated and the results of its initial integration within embedded systems is shown.
4. NOISE DETECTION & ALGORITHM VALIDATION
4.1. Brake Squeal Validation
As stated in paragraph 3, the model was initially trained with data from different vehicles mainly from Mojacar testing vehicles. Data owned by Applus IDIADA from internal testing vehicle was used in order to guarantee the repeatability of the detection in terms of driving style, acquisition system and sensor installation.
After the training phase, the work followed up with validation of squeal data, using a completely independent set of data. Results from four different vehicles were used, and detection using “traditional” techniques (combining subjective driver evaluation and objective detection) with the newly developed algorithm are compared. It is considered that the standard technique (triggers and Boolean conditions) sets the reference value for the detection (100 percent, full detection). Initially a target of maximum 3% acceptable incorrect detection is set.
4.2. Other noises
During the same training phase, the algorithm was also trained for other high frequency noises (multi-frequency squeal and wire brush mainly). However, the occurrence of this specific noises is very low and does not allow to have satisfactory results for the validation (correct detection is currently between 30 and 60 percent and it is not considered as acceptable).
The algorithm proved to be robust and reliable for squeal noise detection. High level of confidence is shown in the validation phase for high frequency noises.
It is clearly shown that the final objective (generating an algorithm which could be embedded in a real time software of a DAQ) is reached.
However, two fields of improvement are shown in this paper and are currently developed in order to have a deeper understanding of the algorithms.
The first improvement fields are related with other noises. Both high frequency events (multi- frequency squeal, wire brush…) but also low frequency noises (creep groan, moan, etc.) should be detected by a robust algorithm. The main limitation in this phase is due to the low occurrence of this noises which could limit the training phase of the algorithm.
The second improvement that is currently under investigation is the correlation between subjective and objective evaluation. The main and final aim of this tool is to be able to assign a rating to the squeal noise which is automatically detected by the system.
About Applus IDIADA
With over 25 years of experience and 2,450 engineers specializing in vehicle development, Applus IDIADA is a leading engineering company providing design, testing, engineering, and homologation services to the automotive industry worldwide.
Applus IDIADA is located in California and Michigan, with further presence in 25 other countries, mainly in Europe and Asia.