Subscribe to Email Updates

sentient-reveals-digitalclone-program-to-improve-wind-turbine-gearboxes_1073_524859_0_7042453_450-228361-edited.jpg

GE Reduces Future Investment in Predix Digitalization Solution: What Went Wrong?

Over the last six years, GE has spent more than $4 billion transforming its 125-year old GE, traditionally known for manufactured products, and set out to turn it into a digital company. GE attributed the shift in business strategy and service to the success of digital platforms like Amazon and Microsoft Corp., according to a recent article published by Reuters.

GE Predix marketed it’s “Digital Windfarm’s” ability to connect wind turbines to computers through big data analytics. The software would predict mechanical failure and in turn reduce costs by analyzing the operational data from the fielded assets, including vibration, temperature, location and other sensor data. The software pulls in the data streams and crunches numbers at its computing centers and alerts the end user to mechanical problems occurring in their machines. The “early alerts” enable service technicians to act, climb uptower and repair the machine before catastrophic failure, saving time and money.

However, last spring GE announced missteps in its “digital transformation” and announced a temporary time out from Predix, citing technical flaws and delays in software capabilities.

The article in Reuters, quoted differences in coding languages and application bugs as detrimental flaws in the company’s ability to continue at the same trajectory.

So, what went wrong the first time?

There are fundamental flaws to big data-only solutions that are available on the market.

Big data works by analyzing anomalies in data. For instance, a spike in temperature or an increase in vibration can be caused by weather events and result in false positives and false negatives. When a turbine technician climbs uptower and is unable to find the problem component or it’s realized too late that the turbine was running with damage, costs of unplanned failures spike, including costs for transportation, labor, expedited shipping etc.

Also, big data fails to predict early-stage bearing failure. Subsurface bearing damage can have a very slow initiation to surface progression, but as the materials degrade, propagation to failure happens fast once it reaches the surface. Cracks spread quickly and wreak havoc on the rest of the machine. A misaligned bearing, leads to secondary damage, unplanned downtime and potential loss of the gearbox.

Materials science provides an additional layer of understanding, which enables a more accurate health prediction of critital components and better insight to predicting damage. Sentient Science’s DigitalClone Live software combines materials science, physics-based modeling, multibody dynamics, and data science to produce a watchlist predicting the mean time to observable damage across three dimensions in time – the “as is,” “to be,” and “future” with life extension actions. The watchlist identifies assets that have components running with damage today, in 0-12 months, 1-5 years and 5+ years.

Life extension can only be effectively achieved with advanced visibility beyond the 6 to 18-month prediction timeframe that big data and traditional condition based monitoring systems provide. To reduce the cost of energy, asset managers need to know the health state of each individual turbine, including the components running with damage, components with early-stage damage and those assets that are healthy.

What does this mean?

There are valuable actions that can be taken when an operator understands the true health state of each of their serialized wind turbines at the critical component level. That data then becomes valuable to multiple layers within an organization.

An operations and maintenance manager needs to know the critical turbines and the assets that will need attention soon. What actions can be taken to get more revenue out of the machine until maintenance can be performed? Will a $7,000 surface treatment give me 6 more months of operation until the low wind season? Or is a physics-based derate a better option, because of the location of the component with damage? What’s the revenue implications of derating that machine?

An asset manager wants to know which assets will fail in the next 1-3 years for optimization on logistics, supply chain and labor. Multiyear budgeting and aftermarket parts forecasting can result in major savings. With watchlist planning, an asset manager can expect to see a $1 million return on investment after the first year, and a $7 million to $10 million return over five years.

The supply chain manager uses the software to reduce inventory carrying costs with greater visibility into gearbox swapouts, uptower replacement forecasts and where in the fleet the needs are located. Additionally, the Buy on Life® feature provides supplier comparisons based on life and life cycle cost. At this level, more than $2.5 million in savings can be achieved.

Risk managers use the software to identify healthy gearbox types in the fleet to negotiate insurance rates and reduce the risk of failed insurance claims through materials science-based root cause analysis reports. An additional $1 million in reduced insurance premiums can add to the value.

These savings added up help to reduce bottom line and decrease the COE, ultimately enabling the company to achieve objectives and revenue goals year over year.

Big Data vs. Small Data

GE, like other big data software solutions, chose to build a horizontal platform strategy for a number of vertical applications that often requires hardware investment in addition to the big data investment. The shear cost of aggregating and then analyzing data to look for anomalies, plus the hardware investments, make these solutions noncompetitive. Customers were looking for solutions that would take their existing data invested over years and provide business insight and reduced cost of energy. Instead of leveraging existing investments, big data becomes another major investment in infrastructure.

DigitalClone Live took a small data, minimalist approach using materials science and existing operational failure data to predict remaining useful life. Pricing for the application is in the hundreds of dollars per asset per year matching customers operational budget constraints and requiring no hardware investment. In this way, Sentient Science captured 10% of the wind market in the first three years of operation and is projecting to have 100,000 wind assets under contract within the next two years. The difference in strategies and technical approaches are clear and dramatic.