In last week’s article on UAVs for Border Patrol, we discussed how re-thinking UAV deployment with...
Developing Predictive Analytics for Pipeline Safety Management
In last week’s article on pipeline safety, we covered the immediate benefits of making an aerial data platform – that leverages UAVS and AI – a core part of your pipeline safety management systems. In this week’s article, we’ll be exploring how you can use the resultant data to craft the ultimate tool for a zero-incident operation – predictive analytics.
In their February 2020 presentation on safety culture, the Pipeline and Hazardous Materials Safety Administration (PHMSA) emphasized the importance of predictive analytics in evolving the industry beyond a bare-minimum compliance approach to safety. Specifically, PHMSA states that operators must work toward a pipeline safety management system that “systematically analyzes safety risk data and performs forward-looking data analytics to identify potential/future problems.”1
However, obtaining these future insights requires a rigorous data sourcing and analysis process. If you’re struggling to understand where to begin, you’ve come to the right place. Here’s a step-by-step guide to how you can use aerial data to build predictive analytics:
Baseline: Starting from a Clean Slate
While a UAV is great for creating a complete snapshot of your pipeline right-of-way (ROW) on a routine basis, AI is what hones-in on the actionable issues amongst the thousands of pictures from each inspection. But, for AI to know what shouldn’t be on your ROW, it first needs to know what should be there. So, at the beginning of an aerial monitoring program, we create a clean set of baseline data to work from. After the first survey of the right-of-way, our team meticulously reviews the imagery and verifies/discards potential anomalies with your integrity management team. With the baseline complete, AI can begin to detect and categorize potential releases, ROW issues, and unauthorized third-party activity. This is human + AI at its finest.
Change Detection: Analyze Your ROW with Laser Precision
Critically, this baseline data is the first step in a process whereby AI compares new and historical data sets to track subtly-developing issues along your ROW – known as change detection. This analysis goes beyond simple pictures, there’s a variety of valuable sensor data you can gather via UAV remote sensing – such as infrared or LiDAR data – that can help you monitor for certain issues. For example, a LiDAR sensor captures precise measurements of the ground surface, allowing for the detection of cracks, erosion, and sinkholes that may signal the kind of ground movement that causes a pipeline rupture.
Critical Points: What Constitutes Actionable Data?
The fruits of this aerial monitoring process are the critical points of data around your pipeline integrity issues. These are the building blocks of an effective predictive model, that you can use to set the thresholds of how issues tend to look along your line.
For example, say during a routine inspection, after months of aerial monitoring of the ROW, the AI system flags a patch of ground directly over the pipeline that is showing irregular discoloration in contrast with prior inspections. The pipeline integrity manager decides it’s worth following up on, and upon close inspection by the maintenance team – it turns out the pipeline is suffering from a corrosion-based pinhole leak that failed to initiate a SCADA alert.
Converging all the relevant pieces of data into a holistic snapshot of the incident, allows us to better anticipate and understand how corrosion incidents may look in the future:
- The precise coordinates of the release along the pipeline
- The season and time of year during which it occurred
- The nature and severity of the corrosion
- The characteristics of the ground discoloration
- SCADA system readings at the time of the incident
- Results from the most recent PIG inspection of the pipeline’s interior
This is the kind of data profile you use to set thresholds for what is likely to constitute an incident along your line. If your team is continually uncovering releases upon an investigation of ground discoloration, that trains the AI system to prioritize similar anomalies during analysis.
Predictive Analytics: Preventing Pipelines Safety Issues
By aggregating these critical points of data, we can fuel the engine of predictive analytics. With the resulting predictive insights, you can identify, assess, monitor, and tackle safety issues on your line before they become critical incidents.
Predictive analytics allow you to:
- Identify Problem Hotspots: Preventing incidents on your line begins with knowing where they are most likely to occur. For instance, say one segment of your pipeline is increasingly subject to unauthorized vehicle traffic. Predictive analytics will flag this is as a problem hotspot, allowing you to start focusing your efforts on how best to prevent them. In this case, the solution may be to install more obvious signage or barriers to entry on the right-of-way.
- Forecast Developing Issues: Many of the safety issues that can plague a pipeline start small and gradually-develop over time into something big. Take the case of construction encroachment, where the site starts out at a safe distance but gradually makes its way to an unsafe distance as the months go by. As AI logs the change from each inspection and starts to form an average rate of change, the predictive analytics model can alert you to when and where this activity is likely to encroach into your ROW. The construction company may not have done anything wrong yet, but it might merit a phone call to give them a friendly reminder of where the limit is.
- Predict Future Incidents: By layering all the valuable sources of data we spoke about earlier and using those to develop thresholds around what leads to accidents, allows a predictive model to make accurate estimates around future incidents. As the system starts to see anomalous and historically-correlated activity, it can alert you to a potential accident before it occurs. Take our corrosion example from earlier, if we’re seeing similar SCADA readings and weather conditions in the same area as our previous corrosion-based pinhole leak – you might want to perform more frequent inspections by the aerial system.
Forward-Looking Data to Achieve Zero-Incident Goals
To develop a pipeline safety management system that achieves your zero-incident goals, you need data that tells you what issues your pipeline faces tomorrow, so you can take action today. There’s no magic button you can hit to access predictive analytics, but by sourcing and analyzing routine, consistent visual data and using it to make sense of your other valuable data sources – you can develop the model that can accurately get you in front of issues on your pipeline.
Have questions about how high-quality aerial data can elevate your organization?
- Pipeline Safety Management Systems – 5 Years of API RP 1173, Pipeline and Hazardous Materials Safety Administration, 2020