Predictive Analytics: For Pipelines or Pipe Dreams? Posted on February 7, 2020 Conventional detect-and-react thinking around pipeline integrity management is causing midstream operators to hit a performance plateau. While new sensor technologies and in-line inspection tools might garner slightly more accurate readings, they still feed into a process that isn’t going to yield operators a dramatic improvement in pipeline integrity. Improving your ability to detect a leak, doesn’t change the fact the leak happened in the first place. To leap forward in operational excellence, operators need an understanding of the trends and potential issues happening on their pipeline before they manifest, otherwise known as predictive analytics. So, what does it take to unlock this futuristic insight? Here’s the truth about predictive analytics, it isn’t just going to “happen”. The machine or software application that will magically spit that kind of data out is still in its infancy. So, is predictive analytics fact or mere science fiction? In the here, now, and foreseeable future – predictive analytics is a process that takes time, effort, and consistency. The predictive part of the term is important because what this boils down to is creating a model for making predictions with a certain sense of probability, based on observations derived from comparisons of historical data. The value for pipeline operators is to gain the insights needed to take preventative action against developing issues on their pipeline, which will drive down costs resulting from leak-initiating damage and broad, unfocused maintenance efforts. To get these actionable insights, a predictive model requires high-quality data to work from. The Predictive Analytics Methodology At first glance, you may imagine SkyX solely exists to make sexy, Porsche-esque drones – and hey, guilty as charged on how nice the drones are – but in fact, everything we do is just a means to high-quality data. In the case of the drone, this unmanned, autonomous system is perfect for performing visual inspection from the skies in an ultra-consistent fashion. With precise programming, the drone can fly the same route, from the same height, and the same angle – every single time. This consistency is important to generating results for meaningful analysis over time. Speaking of analysis, predictive analytics requires data collection to happen regularly over time. While the efficiency of an unmanned, autonomous drone allows for frequent inspection, how do you effectively parse through the sheer volume of raw data coming in from these inspections? At SkyX, we leverage machine-learning algorithms to do the heavy lifting, with a final step of human verification to ensure the findings are legit. To demystify the process that powers predictive analytics, let’s examine how SkyX applies these technologies in the real world to get customers the right data for the job: 1. Baseline Developing predictive analytics begins with establishing baseline information to compare against. With the aerial system, we perform an initial flight to collect a set of visual data. Upon collection, we work with our customers to ensure that the anomalies are identified and worth observing in subsequent flights. Once complete, we have a clean slate with which to compare future inspections. 2. Change Detection To accurately identify how anomalies are changing over time, you need to leverage change detection, which requires consistent visual inspections at a regular cadence. Consistency of inspection is important because accurate comparison of visual data sets requires that image capture is done from the same position and with the same equipment, every time. Regular cadence is necessary because understanding the rate at which anomalies are occurring will inform the probability of future issues. After each inspection, we transfer the raw imagery to a machine-learning algorithm for analysis. With plenty of inspection data sets to meaningfully compare against, the algorithm can leverage change detection to perform laser-sharp analysis. Many of the issues that plague pipelines happen gradually over time – land erosion, corrosion, and other age-related issues – and can be hard to notice at-a-glance, at least until it’s too late. Through change detection, the system flags minor changes that weren’t there before. 3. Critical Points With every inspection, we’re going to generate a lot of raw imagery for analysis. So much so, that it would be overwhelming for a human to handle. To handle the bulk of this workload, we need to leverage computer vision systems for analysis. However, we still need a human to make the final push over the finish line, by verifying the anomalies flagged by the machine. Even today’s most sophisticated AI can’t match the level of certainty a human can provide in analysis, but with the two working in tandem, we can enjoy an efficient process that yields quality results. Here’s how this co-operative effort works in practice: As anomalies feed into the system and are either verified or discarded by a subject matter expert, the system learns what’s important to look for and what to ignore. Verified anomalies from each inspection are going to feed into the model that allows us to set critical points, so we can begin being predictive. 4. Predictive Analytics This final part of the process is all the steps above plus time. As data accumulates from each inspection, we can aggregate when and where anomalies are occurring. With this information, a predictive analytics model can identify the hotspots on your pipeline and estimate how severe they’ll be in the future. Of course, the insights you can derive from this high-quality visual data become much more advanced when you incorporate other layers of data into the process. For instance, when monitoring for corrosion on an above ground pipeline, we know that weather damage plays a significant role in the oxidation process. By checking your visual data against weather forecasts, you further augment the efficacy of your predictive analytics model for a holistic view of this aspect of pipeline integrity. Getting Ahead of the Game Instead of generalizing your maintenance program by performing sporadic inspections with manned aerial or ground inspections, or emergency missions to go see what may have caused a SCADA sensor to trip, you can leverage aerial data solutions like ours at SkyX to get the data you need to get predictive. With the ability to better analyze, identify, assess, and monitor emerging trends and developing issues on your pipeline before they become critical, you will completely redefine the standards of pipeline integrity for the better. Prevent damage. Save money. Improve safety. Reduce labor. Unlocking predictive analytics requires a new mode of thinking about what kind of data is valuable to pipeline integrity management, but making the change will deliver unprecedented results for your organization. Have questions about how high-quality aerial data can elevate your organization? Contact our team to discuss your unique challenges and data requirements.