This is an abridged version of a case study produced and owned by Teradata. For the full copy please visit www.teradata.com.
Today’s car manufacturers face a hyper-competitive global market fuelled by stratospheric development costs; rapidly changing customer preferences; and escalating regulation of safety, environmental and fuel efficiency performance. As a result, the automotive lifecycle is increasingly becoming a global feedback loop in which data produced by vehicles on the road directly shapes the design and manufacture of those still on CAD screens and assembly lines.
Yet finding actionable insights from the flood of field data requires a combination of sophisticated IT infrastructure and data-driven business processes. An organisation must be able to collect and analyse very large data volumes from many disparate sources and then act nimbly on the results. Swedish car maker Volvo Car Corporation has come far along this learning curve.
At Volvo, a system from data warehousing company Teradata integrates product configuration, warranty and vehicle diagnostic data to support technical and business analysis throughout the product lifecycle. The availability of current, quality-controlled data is transforming decision-making processes within the organisation, leading to benefits in quality, warranty costs, customer satisfaction, and bottom line profitability.
From humble beginnings in 1927, the company sold 373,525 cars in 2010 and employs around 20,000 people. Over the past 85 years, the company has been famed for its primary value – safety. Indeed, anti-lock braking systems, energy absorbing bumpers, side-impact protection systems and three-point seatbelts all appeared first in Volvo cars. The company has a goal in place that, by 2020, no one will be killed or injured in a Volvo car.
A second distinguishing Volvo characteristic is its systematic use of operating data from vehicles in the field to improve the quality and performance of those in production and design. In 1999 it began collecting diagnostic read-out (DRO) data as a window into performance and mechanical failure under actual field conditions.
Today’s family car: wired for Data
In today’s cars, sensors and onboard computers perform a wide range of control, monitoring and diagnostic functions in everything from the engine to the braking to the climate control and beyond. Each of these systems generates a diagnostic trouble code (DTC) when it detects some sort of fault.
Trouble codes are typically stored in the engine control units (ECU) until the vehicle goes to a dealership for maintenance. A service technician connects an analyser to the vehicle and reads out the stored codes from the ECU. At a Volvo dealership, the codes are then uploaded to a central database at Volvo headquarters where they form a global reference on all mechanical and electrical failures occurring in all Volvo models.
And that’s not all. Because trouble codes are simple binary signals, they provide no context for the fault conditions they identify. But sensors that measure for fault thresholds are capable of generating incremental measurements, so Volvo equips its vehicles with a variety of data loggers to record a wide range of variable measurements. These include wear factors like catalyst deterioration, and operating parameters such as engine speed and load.
While industry regulators require only four or five of these measurements to be collected, Volvo has extended its systems to collect nearly 400 discrete measurements.
But this means aggregate data volume is rapidly ever-growing. Each vehicle generates 100-150 kB of interpreted data per year. And while it was originally stored in a local data-mart, analysts soon realised that they could not realise the potential value from it in isolation.
Bertil Angtorp, senior business analyst at Volvo, recalls: “We knew there was business value to be gained if we could easily match warranty claims against diagnostic data from actual service records, especially if we could do it without manual integration. We tried to bring the diagnostic data onto the existing warranty platform, but the performance was terrible. Our old system could only handle about 20 jobs at a time. It was a standard joke that when you finally got a response to your database inquiry you’d have forgotten what your question was.”
Volvo analysts began seeking an integrated data warehouse solution, with performance their primary search criteria. In September 2006, Volvo began migrating its data to a new 2-node Teradata Warehouse that went live in July 2007.
The Volvo Cars Data Warehouse brings together data from four primary sources: a system for managing vehicle and hardware specifications, one for managing on-board software specifications, the system that collects vehicle diagnostic data from service centres worldwide, and the warranty claims system. Data access and analysis are enabled through a variety of standard reports and ad hoc analytics, implemented using BI tools and in-house developed applications.
The new warehouse immediately increased the raw data available to Volvo analysts from 364 gigabytes to 1.7 terabytes, and dramatically improved query response times.
A daily fleet mileage calculation that had taken two hours in the previous environment now ran in five minutes and a comprehensive report of diagnostic failure codes by model and year was reduced from two weeks to 15 minutes.
Where performance constraints had restricted access to a handful of users, the new Teradata system extended access to more than 300 users in product design, manufacturing, quality assurance and warranty administration. Where the previous data mart had struggled to process a single query an hour, the new platform completed one a minute.
Immediate cost reduction impact and value return
Operating costs were also reduced by eliminating three single-purpose data marts, and the return on initial project costs amounted to 135 per cent.
“When you consider operation, maintenance, licenses and other resource costs, we’ve reduced infrastructure expense for data management by approximately two thirds,” says Angtorp.
The warehouse also began generating significant new business value through a growing variety of process improvements across multiple operational areas
Volvo analysts had long suspected quality issues in some of the warranty claim data originating in its dealer organisation. Small-scale sample comparisons of diagnostic readout data against warranty claim information for the same vehicles suggested disparity between actual and reported data, particularly regarding mileage. It seemed certain there must be an opportunity to significantly improve reimbursement accuracy if the data quality issues could be resolved.
With all diagnostic readout data – including mileage data read directly from the vehicle ECU – now in the same warehouse with warranty claim data input by dealers, it was a simple matter to compare values and identify sources of high error frequency.
Quality has always been a quantitative discipline at Volvo, with more than 400 members of the design, engineering, manufacturing and quality teams Six Sigma certified. One of the most important impacts of the Volvo Data Warehouse has been vastly improved analytical support for the cross-functional projects these investigators conduct to understand product defects and trace them to their origins in the production process.
“The Volvo Data Warehouse helps us find the cherries in the data,” explains Malte Isaksson, head of Volvo’s Six Sigma organisation. “We’re overwhelmed with data today. We’ve been using computer aided design and manufacturing since the 1970s, and collecting DRO data for more than a decade. Extracting business value from that data depends on our ability to connect chains of events – to see and understand complex relationships that are often hidden by the sheer volume of data coming at us. I think the ability to be able to extract business important information from internal and external data sources will be key to understanding customer behaviour, needs and wants, hence posing a competitive advantage in the ability of doing so. More and more this will become an important aid in decision making.”
Volvo analysts use the Teradata system in a range of sequential investigations aimed at improving quality and customer satisfaction across a product lifecycle that often approaches 20 years. Among the most important focal points are:
Prioritising, targeting, and expediting problem response efforts
The first order of business in analysing large volumes of DRO data is deciding which problems are significant and require high priority response.
“Diagnostic data include many different alarms created by embedded systems within our vehicles,” explains Mikael Krizmanic, senior engine diagnostic engineer. “There’s a hierarchical structure within that alarm data, but it usually isn’t evident. Our first step is correlation mapping to understand the relationships. We do a type of discriminant analysis looking first at mileage and time to identify patterns of occurrence. We put those in dendrograms to identify correlation strength, then group the related alarms. Comparing those groupings gives us our initial prioritisation.”
“Then we calculate failure rates both as a percentage of production and as an absolute number. If you have a fault that occurs at a high frequency but only in a low-production model, that may be much less urgent than one where the failure rate is low but the number of affected vehicles is high.”
Tracing mechanical faults to their root causes
“Correlating and grouping failure alarms also gets you part way to root cause analysis,” Krizmanic continues. “It’s rare that you can get all the way to a root cause from diagnostic data, but you can rule things out and narrow down the possibilities quickly. Once you’ve eliminated the non-contributors you can usually find it pretty quickly, and we usually have between 50 and 100 parameters to work with.”
Modelling failure rates over time
Volvo analysts also calculate predictive failure rates over time using a type of hazard rate analysis based on aging. Each month they look at the number of vehicles that that have reached a certain service age and the number of those that have experienced a particular failure. They do the same calculations the next month and accumulate the results.
“This gives us a cumulative hazard function that tells us how many cars in a given population have experienced a particular failure, and how many are at risk,” Krizmanic explains. “It’s a sort of density function that describes the failure rate over time, and that’s what we use for predictive modelling. It helps us understand which faults will produce large warranty impacts if not addressed systematically.”
Every DTC in the car is analysed this way and can be viewed by model, year, and system.
Correlating mechanical failures with location specific conditions
A car sold in urban China will probably experience different driving conditions and behaviours than one in rural Germany – differences in average vehicle speed, engine load, operating temperature, environmental conditions, time at idle.
“Because we have both DRO error codes and operational log data in the warehouse, we can understand the relationships between geography, patterns of use and mechanical failure,” Krizmanic points out. “A problem may be a high priority for remediation in one geography but not in another.”
Resolving quality issues within the current production run
The faster you can find and fix a design or manufacturing fault the better – for your costs, profitability, and customer relationships. “If we can find and fix a fault in week 26 of a 62-week production run, that’s 36 weeks of production that we don’t have to address in the field,” Krizmanic observes. “It’s a huge advantage both in cost reduction and improved ownership experience.”
Bringing teams together
The warehouse has also been something of a personnel silo breaker. “Product design has become more involved in solving engineering problems,” Angtorp reflects. “They’re now responsible for design throughout the product lifecycle. Our new product development process includes a formal evaluation of past problems and an analysis of opportunities to resolve or avoid those issues in the next design cycle.”
Another benefit has been a significant reduction in time-to-resolution. “Because Volvo now has detailed data on all its cars, they can scope problems more accurately,” explains Teradata industry consultant Torbjorn Rosenquist. “Because they don’t need to integrate that data manually they can act on it faster. And because all the functional teams within Volvo are working with the same data, they can act as one.”
The importance of those capabilities to vehicle quality and safety is self-evident. But they are also proving essential as Volvo pursues a core value with global relevance – environmental sustainability.
Documenting environmental innovation
Beginning in 2008, Volvo has supplied environmentally-optimised configurations of its products. With a lower chassis, aerodynamic underbody panels, minimal engine idling, higher gear ratios, and specially selected wheels, the DRIVe line is designed to maximize fuel efficiency and minimize emissions. All models emit less than the 120g/km limit of carbon dioxide – the Eurozone threshold for special tax treatment. Volvo designers are using the Teradata system to collect and analyze diagnostic information from DRIVe cars in the field to track actual performance against design objectives.
“We want to make sure that our customers actually receive the fuel efficiency performance that we’ve certified,” Mikael explains. “For instance, we know that the start-stop system should engage and stop the engine in 95 percent of all vehicle stops. We can verify that we’re achieving this performance. And in any cases where we don’t reach that target, we can look at the variable data to understand why. We also have instrumentation on board to calculate average fuel consumption, so we collect that data at each service interval.”
Furthermore, charging data from the electrical system is being studied to develop an algorithm for balanced use of engine braking to recharge the battery without overcharging.
Business intelligence development and strategic IT
All future business intelligence initiatives will now also be based on the Teradata platform going forward.
“I would say that today we have only scratched the surface; I don’t think we understand yet, from a business point of view, this tool’s true potential,” says Åke Bengtsson, vice president of quality and customer satisfaction. “I believe that we can better use data to provide early indications. In today’s competitive environment we must be able to act quickly, to reduce the number of steps to an accurate, proactive response.
“Every car we produce with a fault costs the company money. And every minute, hour, and day by which we can expedite a solution saves money for the company. The earlier we can resolve an issue the better it is for the customer and the company. So I think our direction is clear. We’ve recently implemented a hardware and software upgrade that should take us several years ahead, with the performance and capacity we need to really utilise the data we have, and to continue developing new solutions and business opportunities.”
The Teradata system has also earned enthusiastic support from the IT side of Volvo. Jonas Rönnkvist, head of enterprise architecture, recently helped orchestrate its approval as a strategic IT platform within the corporate environment.
“For us, the purpose of a strategic platform is to drive standardisation, commonality, and simplification within the IT landscape,” he explains, “to standardise and streamline the way we build and deliver solutions to the business. The idea is to build and host an extensible set of services or solutions on a core set of reusable platforms, so that we stop building separate solutions (and start) using a single data warehouse in a much more scalable fashion than we have done in the past.
“Every new development project at Volvo Cars now follows a standard governance process, including reviews by my team. If the requirements include data consolidation and integration and the design doesn’t leverage the VDW platform, the project will be ‘red marked’ – it will not be able to proceed without CIO approval.”
For all its impact on vehicle design and quality, Volvo’s data warehouse may have had its greatest impact on the company’s decision-making processes. “Our decision making has become more fact-based,” concludes Bertil Angtorp. “Now, whenever a question arises, people invariably ask ‘what is the data telling us?’ Once we’ve verified the existence of a problem we use the data to determine the scope, to prioritise and scale our response. It helps us make sure that we’re focusing on the things that are most likely to affect the customer experience.”