Deep-learning A.I. will revolutionize the automotive industry

Add bookmark

Peter Els

The motor industry is driven by data collection: Data is collected about component quality from suppliers, system faults are recorded by the On Board Diagnostics (OBD) and stored for later evaluation, not to mention the information collected about customers’ servicing history and preferences stored at the dealer level. 

However this data, often approaching petabytes, is largely inaccessible to users because of the lack of a means to retrieve the relevant data in an understandable format. The amount and complexity of this information simply overload current processing techniques, leaving vast amounts of valuable information untapped and underutilized.

Now, as artificial intelligence (AI) matures and developers explore new use cases for the technology the motor industry is set for an information revolution that will transform every area of the business.

A November 2016 Tractica market intelligence study forecasts that the demand for automotive AI hardware, software, and services will explode from $404 million in 2016 to $14 billion by 2025.

Putting data to work in the field – the impact of AI

AI and machine learning methods are particularly applicable when it comes to powering new insights within the auto industry because the data sets are large, diverse, and change quickly. Applying machine learning to auto industry data is a game-changer, unearthing iterative new connections that can be solve age-old problems.


Take for instance the impact of AI on aftersales service, also known as ‘cognitive predictive maintenance’:

Most manufacturers require customers to return their vehicles to servicing workshops at regular intervals for routine inspection and repair. This is important to ensure the continued reliability of the vehicle and is based on historical service data gained over several model cycles which is then developed into a service schedule mostly based on time and mileage. 

However, this is not ideal and can add unnecessary cost and customer inconvenience when vehicles are laid up for hours undergoing routine servicing that is not necessary at that time. Even vehicles that use on board data to predict servicing tend to fall short as they do not take into consideration all the unique (to specific operating conditions) elements that may be critical to the safe and reliable operation of the vehicle.


Servicing based on the actual requirements of each vehicle would prove more cost effective for the manufacturer while at the same time reducing customer inconvenience. 

This information is largely already available: history of serviceability of wear and tear parts is mostly recorded; vehicle operating conditions can be determined through on board telematics and vehicle tracking; climatic conditions are tracked and can be factored in to extend or reduce a particular service interval; through the OBD the status of critical systems is continually monitored in real time, and any malfunctions recorded as faults.

Putting this data to work, a deep-learning based AI predictive platform would fuse data collected from the OBD, telematics, and fleet history together with customer driving history, customer servicing history, previous unscheduled repairs and recall campaign information.

From this the platform could produce multi-level predictions, such as next scheduled service, components requiring replacement, faults that need repair, service, and parts required, time needed for service and repair, special actions (Authorization for the repair by the manufacturer) and any third party intervention. (e.g. Replacement of glass by a service provider.)

Furthermore, using predictive analytics the driver could be alerted to an imminent service while dealers would also be forewarned of the service so that the necessary workshop time could be scheduled, and the correct servicing and repair parts ordered. Thus the customer gets to spend less time at the dealer; the dealer reduces inventory, and the manufacturer replaces/ services only the required components.

At the dealer level, streamlined forecasting processes would lead to actionable insights by highlighting potential problems that require intervention, or opportunities that could be pursued proactively. For example, based on advanced pattern recognition, the system could predict events such as “out of stock” or overstocked inventory, or could even promote accessories based on a customer’s profile.

AI reduces warranty failure rates, saves lives and cuts costs

Warranty failures and costs are critical to the industry, not only to monitor product quality and reliability and act as an early warning for potential (safety) recall campaigns but also as a tool to identify potential dealer skills deficiencies.

However, the inability to accurately analyze complex data leads to ineffective trend predictions.

Information typically scrutinized in evaluating warranty failures and drawing up forecasts include mileage, date of sale, failed component, failure mode, repairing dealer, component supplier, and costs associated with the repair. Sometimes additional information such as images, video, text or diagnostic file uploads may be included, but these mostly require human review before being transcribed as useful information.

When it comes to complex patterns, traditional predictive methods cannot analyze the full range of structured and unstructured data to predict future failure trends. This is largely as a result of the great differences between components, suppliers and failure modes – as in the Takata airbag recall, the failure is significantly affected by humidity and not related to a ‘batch’ within a specific production run. 

So the problem with the recall is not merely identifying vehicles that operate permanently in humid conditions, but also those that have in the past been exposed to high levels of humidity, and those that travel frequently to these regions.

Using a deep-learning AI platform this complex data, collected from the servicing history, telematics, and failure mode frequency together with parts availability can then produce multi-level forecasts, such as identifying vehicles with similar symptoms’, the probability of failure and therefore priority of the repair, parts availability and dealer location. 

From this, vehicle recalls can be prioritized, inventory directed to the correct dealers and high-risk vehicles recalled as a matter of priority. With limited stock availability and the high probability of failure, this predictive analysis can save lives.

Applying self-learning Artificial Intelligence to carry out forward-looking predictive analysis on all failures also affords manufacturers and suppliers the opportunity to identify potential trends at an early stage, thereby taking corrective action that can save money and lives and safeguard the manufacturer's reputation.

Predicting the failure with self-learning AI

In an industry driven by safety, quality and costs, the failure modes and effects analysis (FMEA) is an important part of the development program, which in turn relies heavily on accurate historical data regarding past performance of the part (or similar part.)

With the wealth of data available from customers, dealers, warranty and part failure analysis, a predictive AI platform would be invaluable in identifying potential future trends resulting from components or systems with inherent flaws. 

It is also not inconceivable that in future, with Industry 4.0, machine settings and processes could automatically be adjusted based on the trends forecast by the predictive AI.

So, while self-learning AI basks in the limelight of the automated car, the real revolution is warming up off-stage. 


  • Martin Prescher; Venture Beat; How AI will play a major role in the auto industry; June 2017;
  • Brian Rashid; Forbes; How AI Pioneers Will Affect The Car Industry, And Why It's A Good Thing; May 2017;
  • Mahbubul Alam; TechCrunch; The top 7 trends in the auto industry for 2017; December 2016;
  • Deloitte; Big data and analytics in the automotive industry. Automotive analytics thought piece; Self-driving perfection is still years away; July 2016;


Company information according to § 5 Telemediengesetz
IQPC Gesellschaft für Management Konferenzen mbH
Address: Friedrichstrasse 94, 10117 Berlin
Tel: 49 (0) 30 20 913 -274
Fax: 49 (0) 30 20 913 240
Registered at: Amtsgericht Charlottenburg, HRB 76720
VAT-Number: DE210454451
Management: Silke Klaudat, Richard A. Worden, Michael R. Worden

Firmeninformationen entsprechend § 5 Telemediengesetz
IQPC Gesellschaft für Management Konferenzen mbH
Adresse: Friedrichstrasse 94, 10117 Berlin
Telefonnummer: 030 20913 -274
Fax: 49 (0) 30 20 913 240
Email Adresse:
Registereintragungen: Amtsgericht Charlottenburg HRB 76720
Umsatzsteuer- Indentifikationsnummer DE210454451
Geschäftsführung: Silke Klaudat, Richard A. Worden, Michael R. Worden