The automotive aftermarket refers to the secondary market of the auto industry and covers retailing and distribution of parts, accessories and chemicals after the sale of the automobile. This US automotive industry is estimated to be worth $318.2B and continues to grow at 16% (as of 2015). The industry is moving towards consolidation as international players and tech companies are trying to get their technology into cars.
As tech companies vie to lay claim on the driver’s seat, our company is disrupting the maintenance and repair aftermarket of these connected vehicles (estimated to be 250 million by 2020). Cyborgs Autocare aims to use predictive maintenance to help create a better network of drivers, vehicle owners, auto mechanics and original equipment manufacturers. Cyborgs Autocare will monitor automotive performance indicators such as to predict when and why a failure is likely to occur and the potential impact of this failure. For a driver, this will help pre-emptively plan their vehicle downtime and automatic maintenance scheduling. Service centers will now be able to optimize spare parts inventory, preempt service schedules and improve overall customer service. For original equipment manufacturers, this will help combat the duplicate parts market better and minimize warranty claims.
As cars continue to become more software dependent, we propose leveraging the stream of performance data these cars are able to transmit in order to develop a predictive maintenance solution that would allow both drivers to be warned in advance of failing equipment as well as dealerships and repair shops to be notified of drivers wanting to come in and equipment needed to service their cars. Using sensors embedded into the car’s system, our product would collect real time data on the engine, exhaust, braking system, airbags, and transmission among other components. This data will then feed into our algorithm that has been trained on historical data to predict the likelihood of needed maintenance. If the likelihood is above a certain threshold, the driver will receive a message on the car dashboard indicating the need and reason to take the car in and the ability to call or schedule an appointment with nearby repair shops or the dealership the car was purchased from.
This solution could also be important for autonomous cars, as it would provide a crucial safety mechanism by automatically directing one of these cars to the nearest repair shop if the algorithm detected any abnormalities.
To demonstrate our product, we would partner with a ride-sharing company, and segment a portion of the ride-share fleet into a number of different pools. The pools would have a comparable make-up in terms of vehicle make, model, mileage, and age. Half of the pools would continue with their maintenance methods as-usual (fixed schedule, and reactive). The other half of the pools would use the predictive model to pre-emptively indicate when maintenance should be performed. Drivers would share all records of maintenance procedures and associated costs. At the end of a certain period of time and number of miles traveled, the results would be compared, looking at the number of failures and associated costs. Variances in the number of miles traveled by the various cars in each pool would be controlled for. The hypothesis is that the pools utilizing the preventative maintenance modeling system will experience lower failure rates and lower maintenance expenses. If successful, this empirical demonstration would help promote the merits of the solution.
For collecting the data, we would consider two strategies: using publicly available historical data and collecting data directly from the vehicles. One example of publicly available information is the NHTSA complaint database (https://www-odi.nhtsa.dot.gov/owners/SearchSafetyIssues), which contains structured information and text complaints on vehicles going back to the 1950’s. We would use natural language processing techniques to extract specific part failures from the text information. In addition, by 2020, more than 250 million cars will be connected to the Internet (http://www.gartner.com/newsroom/id/2970017). Assuming we are able to negotiate a partnership with vehicle manufacturers, we would be able to continuously collect information logged on parts as it is sent from the vehicle, and vastly increase the our database.
In terms of creating predictive models for maintenance, we would combine several techniques to develop an overall picture of vehicle health. One method would rely on survival analysis models for individual parts, which estimate the expected duration of time until a part fails. In addition, we would create a Markov chain model that would tell us what part is likely to fail next given existing information and previous failure history. These two models can be combined to create an overall vehicle health index. In addition, we are confined in this approach because these models have been proven to work in other industries such as oil and gas (http://www.ospreydata.com/architecture/).
