At SparkCognition, this is a quandary we hear every day: “We have all this data — now what can we do with it?” Luckily, it’s also a problem our suite of solutions can assist with.
Challenge 1: Massive quantities of data
Although sensors have provided huge amounts of data, it often sits at the point of collection, remaining unorganized and unusable on a larger scale. Applying advanced analytics in-house is generally not an option — it is both time-consuming and challenging to assemble the right team of experts who will then need to re-invent the wheel to create a proprietary system. Companies are stuck needing to show ROI on endpoint investments but without the tools to do so.
SparkCognition’s analytics solution, SparkPredict, uses cognitive analytics to turn data lakes into actionable insights on asset behavior, which can lead to increased productive time.
Knowing the physical responses of how an asset behaves is critical to optimizing its use and identifying anomalies that could indicate a breakdown. SparkPredict can aggregate data from thousands of sensors and find nonlinear correlations that point to how parts of the asset are working together. SparkPredict also analyzes data during transient events such as startups and coast-downs, a period of time when most issues first appear yet is not accessible to traditional static models and pattern recognition tools. This provides unprecedented visibility into asset performance.
Our unique automated model-building algorithms can also identify anomalies, without prior knowledge, that other monitoring systems focusing on fewer variables will miss. By identifying potential failures earlier, companies have the opportunity to streamline and prioritize maintenance schedules, instead of waiting until a problem occurs. Planning for maintenance also allows downtime to be minimized, keeping machines productive.
A study by Dartmouth’s Tuck Business school this year found that the increased insight into turbine system performance can provide three main advantages: reducing downtime (by 50%), decreasing response time to failure (by 25%), and reducing catastrophic incidents (by 35%). The study estimated the increase in production would amount to $1.3M annually for a combined cycle gas turbine power plant.
In a recent deployment, SparkPredict was running on a fleet of new turbines for a top three energy provider. Within four months of deployment, SparkPredict identified a manufacturing defect that went undetected by traditional monitoring systems lacking a holistic view of the turbine operations. Had service not been performed, this defect would have resulted in catastrophic damage to a $100M asset.
Challenge 2: Capturing information outside of sensors
While the dataset from connected devices is incredibly large, it still does not capture all the relevant information to operations. Sensors alone can’t accomplish this: Valuable information lies in maintenance manuals and service notes that are difficult to incorporate into the body of knowledge.
While SparkPredict can provide alerts on anomalies, natural language processing carries the potential to provide service recommendations. Our solution, DeepNLP, uses artificial intelligence to process unstructured data like text files, tables, and PDF manuals and convert them into a searchable database. When an alert arises from SparkPredict, DeepNLP can then find the corresponding issue within its repository and suggest action.
Challenge 3: Cybersecurity
Connectivity at endpoints opens them up to cyber attacks. While in the past, equipment sat protected behind barbed wire and fences, a more sophisticated solution is required today. Malware continually mutates and evolves, and preventing outages requires a cybersecurity system that can keep up.
SparkCognition provides end to end cyber protection with the addition of DeepArmor, which uses automated model-building algorithms to anticipate malware. Cybersecurity and cognitive analytics combine to provide a complete solution suite and peace of mind that assets are running without interference and at lower risk.
Artificial intelligence can help utilities companies truly maximize return on machine data, sensor data, and process data by providing deep insights on asset performance and keeping critical systems secure. After all, the data is being collected — why not make the best use of it?