Predicting Remaining Useful Life (RUL) of Automotive Components using AI/ML technologies

The automobile industry is witnessing a paradigm shift in technology. Vehicles have become smarter over the past few decades. Technologies like Big Data, Data Analytics, Artificial Intelligence (AI), Machine Learning (ML) are helping drive this trend.

Starting from 3-4 ECUs per vehicle, modern cars have as many as 150+ ECUs per vehicle¹ to offer a wide range of features.

In parallel, the need for CASE (Connected, Autonomous, Shared, Electrified) vehicles is leading to an architectural transformation. This transformation comes on the backdrop of humongous data generated by vehicles in real time.

Do you know that a connected vehicle generates about 25 GB of data per hour²?

This data can offer OEMs, the power to see the unseen. Data science & data engineering techniques and technologies can help OEMs generate trusted insights & make efficient decisions which in turn can lead to significant cost saving and open new avenues of revenue generation.

One such avenue where this data can potentially generate actionable insights is Predictive Maintenance. Predictive maintenance is the potential to apprehend the failure of components proactively & act upon it in advance.

Table 1: Input data requirement

 

While both preventive & predictive maintenance are proactive in nature, they differ in many ways. Let us understand in detail:

Until recently, preventive maintenance has been the key to upkeep the vehicles & reduce the possibilities of catastrophic events.

With more than 100 million lines of code³ in a single vehicle, a modern vehicle is as complex as an aircraft! In such an ecosystem of complex interconnections & dependencies, there is a human limitation to preventive maintenance techniques.

As the complexity increases with the increase in number of ECUs in a vehicle⁴ , the associated risks and life hazards increase exponentially. Therefore, the importance of predictive maintenance and Remaining Useful Life (RUL) prediction is going up. Preventive maintenance methods are often inadequate to stop a failure/event, like the failure of brake pad or gear box. That is where predictive maintenance in combination with preventive maintenance can change the story altogether.

According to a McKinsey report⁵ , manufacturers can reduce maintenance costs by 18 to 25 percent by deploying analytical techniques.

The Remaining Useful Life (RUL) of a component is a method of predictive maintenance using machine learning & AI to tell us how long the component can be used before the potential failure occurs. The mechanical systems within a vehicle are very complex and hence establishing a linear relationship to the degradation process for every component is not possible. That is where data driven methods come to one’s rescue. Current RUL methods are data driven methods of prognosis.

RUL enables OEMs and component manufacturers to improve customer satisfaction, predict failures before they occur & improve product quality by applying AI/ML technology. It finds great use in dynamic supply chain management of spare parts of a vehicle/machine. It helps maximize the usage of parts which differs from vehicle to vehicle and minimize the downtime costs associated with unscheduled maintenance activities. It has varied applications in anomaly detection, event prediction, and data engineering.

Predictive maintenance is particularly very useful for OEMs to:

• Increase vehicle uptime
• Reduce warranty & recall costs
• Reduce quality issues
• Improve customer services
• Handle Dynamic supply chain management

One of the major factors affecting the choice of buyers while purchasing a vehicle is the quality and ease of aftersales service network. Predictive maintenance can play a crucial role here.

In the past, the prohibitive costs of implementing the architecture on a mass scale limited the technology development in this space. But with ever reducing costs of cloud⁶ and supporting infrastructure and the embedded connectivity in modern vehicles combined with rising demand for proactive maintenance, RUL is now economically viable, scalable, and production-ready.

The below table shows the trend of computing costs over the years:

Source: https://appdevelopermagazine.com/why-the-cost-of-cloud-computing-is-dropping-dramatically/

KPITs RUL prediction model, an advanced machine learning platform, is a proven solution already in production for the past 2 years. It is easily deployable as Software as a Service (SaaS) solution. With a deep domain knowledge across various technologies in automobile industry combined with extensive research & analysis by our experts, KPIT is one of the leaders in the RUL space.

KPIT has developed proven predictive maintenance algorithms for front/rear brakes, engine oil, Li-Ion battery, and spark plugs. Given the wide application & advantages it brings, we plan to expand this library to cover many more use cases like brake fluid, coolant, lead-acid battery, SCR catalyst etc.

KPIT’s flexible architecture of RUL solution is scalable, cloud agnostic, and easily adaptable to multiple applications. Our predictive maintenance solution can be integrated with any existing connected vehicle platform or existing application via an API (Application Programming Interface). The API then pulls the relevant parameters from the existing system, run the algorithms on data and push the insights via API in required formats or dashboards.

The RUL algorithm fetches data from the sensors via CAN bus within a vehicle. It could be location, temperature of the engine oil, pressure of oil through the wall, driver behavior, speed, DTCs etc. The data is collected by TCU (Telematics Control Unit) and pushed to the cloud where it is then processed and analyzed by the RUL algorithm to create useful insights. TCU also provides data on driving patterns like acceleration, braking, turns, speed etc. which can tell us about the quality of driving. Using timestamp & location it is possible to pull information about weather, ambient conditions, road conditions etc. using 3^rd party APIs.

Therefore, RUL combines data on 4 fronts:

• Driver behavior/quality of driving
• Vehicle parameters
• Environmental parameter
• Location parameters

The component-based algorithms fetch the required data to produce useful insights to predict failures, remaining life of a component, AI based diagnostics etc.

It creates insights in the form of number of miles/ kilometers left before the failure or number of hours left before the failure.

RUL can intake data from any source if the relevant parameters are made available. Data can be pulled from the existing data lake, or it might as well be collected by ELDs (Electronic Logging Devices) or Telematics Units (TU) with a cellular connectivity or Wi-Fi connectivity.

RUL solution can be easily integrated with guided vehicle diagnostics to enable a real time analysis of vehicle health.

When combined with OTA updates, it can send updates to vehicles based on a bunch of machine learning algorithms.

Given the multitude of benefits it brings to the table, predictive maintenance in conjunction with preventive maintenance is the way forward in automotive industry.

We at KPIT, are all geared up, and are well positioned to help you leap ahead of the times for a cleaner, smarter, and safer world.

References

1. https://developmentuat.kpit.com/popup_questions/average-number-of-ecus-a-high-end-luxury-car-have/

2. https://www2.deloitte.com/us/en/pages/advisory/articles/connected-vehicles-shift-an-industry.html

3. https://cmte.ieee.org/futuredirections/2016/01/13/guess-what-requires-150-million-lines-of-code/

4. https://spectrum.ieee.org/software-eating-car

5. https://www.mckinsey.com/business-functions/operations/our-insights/digitally-enabled-reliability-beyond-
predictive-maintenance

6. https://appdevelopermagazine.com/why-the-cost-of-cloud-computing-is-dropping-dramatically/