Primary Use Cases
1. Moving From Calendar-Based Intervals to Real-Time Predictive Maintenance
At Oshkosh Corporation, engineering teams use Dataiku to create predictive maintenance analytics that save their customers time and money by keeping vehicles on the job and out of the shop. A great example of this is a routine oil change. Similar to a car, most heavy-duty vehicles typically use calendar-based scheduled intervals (e.g., every six months), without considering the historical operations and current oil quality. Through vehicle telematics data and condition-based maintenance (CBM) methods, oil degradation can be monitored and modeled to accurately predict a remaining useful life and prescribe appropriate actions. By triggering an oil change only when needed, customers avoid unnecessary maintenance actions that result in inflated materials costs and reduced vehicle availability.
The actionable insights produced through CBM analytics projects in Dataiku have shown tremendous potential to disrupt the current maintenance paradigm for vehicle fleets in many markets. Early pilot program results show a potential 2-3x improvement in required maintenance intervals, which translates to more than a 50% reduction in downtime and support costs over the life of the vehicle. If extrapolated over the fleets of similar Oshkosh vehicles in operation around the world, this could result in thousands of labor hours and millions of dollars saved for their customers.
There are additional benefits to these analytical solutions, too. A clear win is the increased environmental sustainability via reduced use of hazardous materials such as oil — literally saving thousands of barrels of oil in the previous example. A more subtle benefit comes from reducing additional vehicle faults and failures that are often accidentally induced by performing maintenance actions, which would mean additional rework, time, and costs. Lastly, indicators from CBM models, operationalized through Dataiku, keep operators from running their vehicles to critical failures while in use, preventing additional incidents that could be potentially dangerous and much more costly to repair.
2. Enhancing Product Intelligence With Edge Computing
Another way that Oshkosh is building out a forward-thinking data science practice is through its product intelligence initiative. With edge computing, ML is embedded directly into the vehicle to generate advanced analytics and model insights locally in real time without the need for (and connection to) a centralized cloud. This is best seen in their commercial vehicle markets where new technologies offer competitive advantages for both the manufacturer and the owner/operator through value-added innovations and increased product efficiencies.
For example, consider the world of concrete. An average business might have a fleet of vehicles and several cement plants around a general service area. Over a typical workday, a single driver/truck will be dispatched many times to various locations, and each delivery contains many routine operations, such as loading and pouring product, cleaning equipment, driving around, and so on. However, since every dispatch can vary greatly, the timing of these operations and total delivery are often hard to estimate ahead of time, which can make logistical fleet coordination difficult and inefficient.
Using Dataiku, Oshkosh rapidly developed efficient edge solutions to automatically identify and report a vehicle’s current operational status. Not only does this give the dispatcher more time to utilize more accurate information for better decisions, it also frees the driver from tedious additional duties and possible safety distractions.
Altogether, these small savings compound over time and have a multiplier effect when scaled over the entire fleet. A few minutes per trip becomes many hours over the fleet. This is significant time that directly benefits the bottom line, either through increased capacity and revenue potential or by way of decreased downtime and costs through more ability to service vehicles without impacting operations and further mitigating more costly breakdowns that can cause extended downtime and service interruptions in the future.
3. Freeing Up Manpower and Transforming Product-Specific Analytics
In addition to customer-facing data science and ML initiatives, the team also performs a variety of data work for internal customers, such as dashboarding and automated reporting. At the heart of these tasks is reliable data engineering and robust descriptive analytics — both of which are facilitated by low-code solutions in Dataiku that empower less experienced analysts to create high quality, reproducible, production-ready results.
For example, Oshkosh must meet strict performance requirements when testing new vehicles, and the data is carefully analyzed internally and routinely audited for compliance. Before the team’s involvement, this was an extremely manual process involving multiple data transfers between disparate systems and siloed workers.
By creating a single unified workflow in a Dataiku project, all parts of the process were linked together through an observable data lineage with embedded documentation that made collaboration and process improvement efficient and effective. This included:
- Manual data prep (cleaning, enrichment, etc.) was replaced with visual prep recipes, and Excel/VBA macros were recoded as Python script recipes.
- The deliverable report to management (Excel charts in PowerPoint slides) was automated using Python to seamlessly generate the charts and slides exactly as expected.
- The data product was then synced to their Hadoop environment where self-serve BI dashboards automatically update for end users to quickly identify and investigate issues and trends that may have otherwise gone unnoticed.
In all, this entire process that used to take hours can now be done automatically in a matter of minutes whenever the customer desires.
This project highlights some of the major recurring strengths of Dataiku for Oshkosh: transparency, repeatability, and maintainability across the team — all of which minimize single-point failures and promote collaboration and innovation. These types of projects are relatively low-hanging fruit across the organization, but their benefits are truly game-changing.
In just the first year, these operationalized solutions are already saving tens of hours of employee time per week that can be better spent on more meaningful parts of their job. In addition, these are excellent opportunities to promote a culture of data literacy and data stewardship, as well as identify potential citizen analysts from within that already have strong domain expertise and a vested interest in improving their work. Dataiku allows the team to quickly onboard new data workers, leverage existing projects/results, and scale data-driven analytics across the enterprise.
You need collaboration to do data science effectively in the first place. Dataiku facilitates and enables collaboration much better than anything else I’ve seen. It used to just be technical experts and data scientists, but now analysts and management can be involved side-by-side in the process.”
Dr. Michael Schuh, Chief Data Scientist, Oshkosh Engineering