This post was released as part of the Data Science Blogathon.
Introduction
The auto industry has mobilized the economy for decades. In the automotive industry, the value chain begins in Inbound Logistics (raw material, stock) Production (Operation), Marketing and sales, Maintenance (Service) given the design of the automobile, vendors, the procedure is ready.
Even though data science is used at all levels of this value chain, such as schedule optimization, transport optimization, inventory management, optimization of the workforce, process optimization and process control, targeting the right audience for the right car segment, today I will describe more in the warranty or maintenance period analysis use case that I have worked on.
Why warranty analysis?
After sale, cars get after-sales services from dealers. A warranty analysis is based primarily on the data collected from those services, claims over a specified period.
In warranty analysis, Gamma distribution is typically observed, Weibull or lognormal for product failure throughout the period.
What does the collateral analysis data look like and how do we fit the distribution to the data?
A collateral analysis is the analysis of elapsed time data until event / failure. In our example, the individual part is tracked from the time of the car sold until its breakdown.
As in typical model construction, we divide the data into test and train data sets. With training data, we first estimate the parameters of the distribution and then, using test data, we see if the fitted model works well with that data or not.
Range: The probability density function (pdf) of the Gamma distribution can be expressed as:
Gamma distribution
with α> 0 y β> 0 yx ≥0.
This also means that Ihas two parameters, alfa– that represents the shape parameter, Y beta- that represents the scale parameter, and its support consists only of positive values.
In the Gamma distribution, using all xi and using maximum likelihood estimates, I calculated the α and β for each element (spare part in analysis). But I observed that the Gamma distribution when fitted to the test data, the actual time to fail was not close to the predicted time to fail by model for that item. Hence I discarded the gamma distribution approaches.
Weibull: The probability density function (pdf) of the Weibull distribution can be expressed as:
Weibull distribution
with α> 0 y β> 0 Y 0 This means it has two parameters, alfa– that represents the shape parameter, Y beta- that represents the scale parameter, and its support consists only of positive values. The scenarios in which the car is driven, driver's driving skills, the age of the piece are some of the parameters of which some can be measured and others are qualitative. Y, therefore, post failure modeling needs to be done on a large amount of data. The rest of the model will have a bias from the modeling data and the precision of the model will decrease over the implementation period. / implementation.
Challenges: Typical challenges would be data accuracy and timeliness, the length of the historical data, the number of parts to be modeled and the selection of the model based on the accuracy of the test data. In my analysis, I had the last two challenges due to reasons like the car will have hundreds of items and the model that best fits one item may not fit another item well.
Solution: There are many research papers that have explained this problem, They mentioned that the distribution of three is a way to go. To the problem, was solving that the Weibull distribution fit all the items in the piece well. Once i'm fit, the test data should support the chosen model and once this is done, I developed the procedure for each item to be modeled, and its scoring function is stored in such a way that the failure forecast per item is generated periodically.
Once this solution is implemented in the business procedure, every month, manufacturers will have an estimate of which posts and their count should be ready for potential warranty claims. This makes it possible to save on customer waiting time for parts to be available., better inventory management, Lowering the cost of unnecessary storage of non-essential parts and also verifying fraudulent warranty claims.
Conclution: For all manufacturers, warranty tests can fix the following problems:
- Identify complaint patterns (seasonally, the mileage, etc.)
- Predict the expected number of claims or the cost of claims
- Predict fraudulent claims (a particular dealer always files warranty claims, the useful life of a post is very long, but those
parts come in often for a warranty claim, etc.) - Investigate the association between different types of claims (the two spare parts used at the same time has the
same failure time) - Identify problems before they become serious and stop working.
- What-if analysis, as if we increase the mileage, which will affect warranty costs.
The benefits of collateral analysis are:
- Higher customer satisfaction, product quality and brand status.
- Tremendous impact on bottom line due to early problem identification
- Optimized guarantee policies for maximum financial performance
- Increase the efficiency of support logistics, such as optimal storage of spare parts or deployment of technicians.
It can also help answer questions like:
- If we can increase the product warranty from 3 years to 6, if the data supports the product failure occurs only after 10 years. If a competitor has increased their warranty and if we adopt the same, How many additional warranty costs will we incur? If we don't, How much revenue will we lose from the decline in market share?
- Given a new product with no historical data, Should we play it safe and offer only a one year warranty, or we can offer a three-year warranty to boost our brand awareness?
This post summarizes the background, the response approach and benefits of collateral analysis, which is a very large industry due to increased sales and competition in the automobile market.
The media shown in this post is not the property of DataPeaker and is used at the author's discretion.
