AI/ML applications needed for Automotive NVH refinement

Various scientific tools such as transfer path analysis, sound quality definitions, Jury evaluation, driving simulators, etc. have been in use for automotive NVH refinements. But still there are some areas where AI/ML can make the whole development faster or more efficient or effective.

For condition monitoring, the vibration-based AI/ML models are ready to analyse complex patterns of bearing vibrations in order to correlate them to specific faults with inner or outer race or rolling elements mainly due to wear or ageing.

Let us note that traditional methods such as Order Tracking or FFT or Cepstrum analysis are not fully effective in the above case when it is observed that faulty or worn-out bearings introduce often random &/or abnormal frequency components or non-repeating patterns in their vibration signals varying with time. Past large recorded vibrations data of various bearing units analysed by AI/ML algorithms attempts a better correlation with the field failures.

Another example on the same line could be piston-slap identification from measured cylinder-wall vibrations & near-engine noise under various load and speed conditions.

Of course, it needs first a lot of test data of piston-bore clearance and corresponding vibrations & sound pressure signals to train the AI/ML model.

For passenger cars, In-cab quietness need be assured along with cost and weight optimization of acoustic insulation material. For this accurate prediction of all noise sources is essential. Though a no. of transfer path analysis tools &/or CAE based statistical energy analysis methods are available, it is difficult to capture all the noise sources or their ATFs (acoustic transfer paths) 100% right unless huge resources, cost and time of development will be made available.

In such a case, AI/ML model, after it has been trained to see, among a dozen of benchmarking cars, a common pattern of contribution of road /tyre, power-train & wind noise components to the total In-cab sound level, can be used to predict the in-cab noise of a similar kind of a future vehicle on the similar roads with > 90% accuracy.

For both human-driven and Autonomous vehicles, AI systems can learn from individual driver behaviours and preferences to personalize their In-cab sound quality.

By understanding how passengers have NVH perception, engineers can develop adaptive control systems that adjust vehicle parameters in real-time to improve comfort.

for Electric Buses, the cooling Fans for batteries, inverters, motor generate a large sound once they are driven into hot areas. The same is true for HVAC compressors which create high noise levels at top speeds under higher thermal loading which is fluctuating as per customer's driving pattern.

AI/ML algorithms can analyse real-time data from temperature sensors, power consumption, etc. to dynamically adjust the speed & duration of cooling fans and HVAC compressors and thereby controlling sound of these EVs both outside & inside the cabin.

Thus, a future NVH Engineer is likely to rely more on AI/ML tools to make the vehicles more refined! But to train them, the input data generation of various transfer paths of structure-borne and air-borne noise correlating well with the human perception on multiple existing vehicles of will be required.