The age of networked production brings with it numerous new challenges. One question that companies in the industry in general and in the automotive sector, in particular, are faced with is that of lucrative data science use cases. How is added value effectively generated from the large amounts of data? Predictive maintenance ( Predictive Maintenance ) has been one of the new standards developed in the industry in recent years. Many car manufacturers and manufacturing suppliers have since benefited from data-based maintenance. At the center of predictive maintenance is the concept of data mining.
Table of Contents
In A Nutshell: This Is Data Mining
The term data mining is part of big data. The experimental methods – partly fully automated and somewhat semi-automated- gained from large amounts of data can be subsumed under data mining. Data mining aims to promote (or “mine”) dependencies, regularities, and patterns in otherwise disjointed or unstructured raw data. Data mining methods are statistical processes that allow the data to be:
Depending on the use case, these methods can or must also be combined. A range of techniques is subsumed under data mining that allows us to deal with data in a meaningful and profitable manner. Large amounts of data are generated in industry, especially in the context of monitoring. In the course of their evaluation, new business areas and models can emerge. For example, in the automotive sector, fleet analyzes can be carried out, which means that customers can be offered a completely new service model. If conspicuous patterns in the data indicate a possible defect in a component, it can be replaced before it causes damage.
Predictive Maintenance In The Automotive Sector
Predictive maintenance is about the analysis of large amounts of sensor data. In this case, we speak of data mining when the specific question that a data science project should about already exists. In the case of predictive maintenance, this is: “When is a machine such as a car engine or even just individual parts likely to fail?” Based on these forecasts, maintenance can be planned before a part fails.
In concrete terms, this means that sensors must first be placed in many different places in the engine compartment and the vehicle. Often specific measured values are even collected several times to rule out that a sensor delivers incorrect values or that measured values do not represent the actual status quo. This creates hundreds of gigabytes of data that are analyzed for appropriate patterns.
In predictive maintenance, there are two data mining tasks: The first step is to define a normal range – i.e., the parameters within which the error-free functioning of a machine or motor can be guaranteed. Significantly different values are measured for this purpose:
- Fluid levels
The more measured values are available from other areas, the more dependencies on different variables can be visible during data mining.
In a second step, a search is done for patterns that indicate that a particular part within the engine could cause damage or fail. Comparing the standard and deviating samples provides information on whether a component will soon have to be serviced. The specialty of predictive maintenance: the longer a predictive maintenance model is used in practice, the better it gets. If a particular pattern is recognized in one case, it can be transferred to the entire fleet. In addition, the knowledge gained here can work back into production, where the causes of errors can be eliminated quickly and at an early stage.
Data Mining In Practice: Predictive Car Maintenance
For one of our customers in the automotive industry, the challenge was to identify vehicles with a possible defect at an early stage before errors occur. In this way, warranty costs should be effectively reduced or avoided entirely. The solution consisted of creating a forecast model that used data mining methods to evaluate various measured value data, the master data of the vehicles, and diagnostic data. During the project, the data first had to be analyzed to define the regular operation. The deviations from the norm only became visible when the current data was analyzed. The advantage of this method is knowing that a defect will occur but which component is precisely the cause.
As a result, this forecast model was able to identify 75% of the vehicles affected by errors in advance. As a result, the testing costs and time-consuming product recalls were avoided mainly completely. As a result, the warranty costs were reduced by over 50%. Measures like these increase customer satisfaction and the reputation of the brand at the same time.
The Future Potential Of Data Mining
In this or a similar way, data mining is at the center of numerous use cases in the field of networked, industrial production, or networked products. The use case of predictive car maintenance briefly presented here shows how great the potential of data mining is. This creates advantages both for the industry and its customers or consumers, who benefit from more reliable products and more service offers. In the future, data-driven business models and data science projects will offer companies the opportunity to continuously improve the quality of their products and enormous growth opportunities.