How energy customers are unlocking the potential of generative AI for Asset Performance Management and methane leak detection

Introduction

Industrial Asset Performance Management (APM) has experienced a significant transformation over the past three decades. Organizations are shifting away from rigid, time-based maintenance schedules toward predictive maintenance, which enables them to optimize cost, enhance performance, and mitigate risk by anticipating equipment failures before they occur.

Today, energy companies are leading the adoption of new technologies in this field. This post is about how a leading Oil and Gas company is using innovative APM solutions that use artificial intelligence/machine learning (AI/ML) models and generative AI along with Industry 4.0 technology to offer online remote monitoring and real-time analytics for critical assets and “balance-of-plant.” This demonstrates how they unlock unprecedented levels of efficiency and innovation, thereby transforming traditional processes and driving significant business growth.

Predictive AI and physics-based models use structured data from machines, sensors, and operational processes to optimize performance outcomes, providing real-time alerts and minimizing unplanned downtime in a predictive maintenance framework. However, these technologies currently underuse the extensive repository of tacit knowledge contained within unstructured data sources, such as maintenance histories in work order management systems, technical documentation, and established best practice protocols. With the advent of generative AI, these APM solutions now offer an ability to make user interaction with their solutions seamless, conversational, and highly actionable.

About Shoreline AI

Shoreline AI’s plug-and-play APM delivers breakthrough simplicity and cost efficiencies. This cloud-native approach requires no new CapEx, on-site experts, or data scientists, operationalizing in days and delivering powerful machine-specific analytics. This highly secure, 100% subscription approach creates unprecedented APM economics and scales for new Environmental, Social, and Corporate Governance (ESG) applications such as leak detection, maximizing renewable assets, improving energy efficiency, and emissions monitoring.

Figure 1: Shoreline AI’s cloud-managed APM platform.

Architecture

The following is a high level architectural diagram of Shoreline AI’s solution on Amazon Web Services (AWS):

Figure 2: Architecture diagram for Shoreline AI’s solution on AWS.

The high-level architecture (see the preceding figure) for the data ingestion and processing into Shoreline AI’s AWS environment consists of several AWS services. Data is sent to AWS IoT Core on a schedule. When an anomaly is detected, AWS IoT Core handles device authentication/authorization and data encryption, and then transfers to a tenant-specific environment in the Shoreline AI environment. The edge devices use FreeRTOS as a real-time operating system, which includes libraries for security, and over-the-air (OTA) upgrades to the firmware. Data can also be ingested from the users’ existing industrial systems (SCADA, data historians, PLCs, etc.) or third-party sensors. The data received by AWS IoT Core triggers AWS Lambda for processing and transformation through AWS IoT Core rules. Structured data is stored in Amazon Aurora, which is a relational database service compatible with MySQL and PostgreSQL, while unstructured data and asset manuals are stored in Amazon Simple Storage Service (Amazon S3). Shoreline AI also uses Amazon SageMaker for training its ML models. This combination of ML models with physics-based models enables the auto-configuration of assets. This auto-configuration draws from Shoreline AI’s proprietary pre-built library of more than 30,000 assets. Initial machine baselines established from the auto-configuration process provide insights in days, not months.

Amazon Bedrock is used to analyze and unlock value from structured and unstructured data. This, in combination with the latest in Large Language Models (LLMs), creates a richer experience for Management and Field Technicians, essentially creating an “Enhanced APM”.

Customer impact

Shoreline AI has implemented this solution for a Top 5 American multinational energy company with 200B+ USD annual revenue and with 1000s of locations in the United States.

The organization uses a diverse range of machines in the production process, such as reciprocating compressors, engines, motors, tankers, pipeline, tanks, and valves. By using predictive maintenance strategies in an APM solution, the customer aimed to enhance equipment reliability and uptime, optimize maintenance costs, and improve safety and compliance of its assets. Additionally, the customer aimed to facilitate data-driven decision-making, thereby reducing operational costs and improving overall efficiency.

APM Solution
Prior to deploying Shoreline APM, our customers were experiencing more than a twofold increase in downtime costs over two years, with the cost of an hour’s downtime now exceeding $500,000.

Using Shoreline AI’s solution, the customer has not only minimized unplanned downtime, but also reduced environmental cleanups due to equipment failures, saving millions in potential production losses and costly repairs.

The deployment of the APM solution on various assets across 30+ sites helps the customer gain real-time visibility, automate inspections, and enable early detection of asset anomalies such as mechanical looseness, misalignments, imbalance, valve conditions, bearings issues, and much more.

This solution offers highly reliable and accurate insights that not only analyze the raw data of machine parameters coming from Shoreline AI’s proprietary Smart Sensors, but also combine process data from historians and SCADA systems. This solution uses physics models and AI/ML to understand the machine behavior and predict failure events with a high probability.

Figure 3: Customer’s assets being monitored by Shoreline AI. The predictive maintenance solution was deployed to automate the detection of anomalies in the machines’ operating conditions. Smart sensors shown in the photo in the blue circle.

Figure 4: APM solution images for a natural gas compressor.

Emissions Leak Detection solution
The customer also has several assets that are at risk for fugitive methane and Volatile Organic Compounds (VOC) leaks. These include oil storage tanks, wellheads and associated pumps, pressure release valves and compressors, pipelines, processing facilities, and gathering and boosting stations.

The same combination of Smart Sensors and the cloud that is deployed for APM is also extended to offer Emission Leak Detection solutions. However, the AI/ML models used for monitoring for methane and VOC emissions are different from those used for APM. These models are optimized for enabling the early detection of leaks with precise location details. From the customer’s perspective, they are accessing the same user interface for both applications.

Shoreline AI’s Emission Leak Detection solution generated immediate notifications and the precise location of leaks and emissions such as methane and VOCs. This solved a critical challenge for the company with such high volumes of industrial fuel storage tanks and upstream Oil and Gas assets.

All of this has enabled the company to take rapid steps to reduce fugitive emissions, helping them comply with stringent environmental regulations and, more importantly, reduce their impact on the planet.

Figure 5: Shoreline AI’s real-time monitoring of Oil and Gas storage tanks to detect methane leaks.

Figure 6: Methane leak detection solution.

Enhancing industrial APM with generative AI
The APM Virtual Assistant application (an extension of Shoreline AI’s APM solution) offers improved knowledge management, intelligent queries, and dynamic analytics. This enables management, plant operators, and supervisors for the customer to use a mobile chatbot to instantly query and analyze critical equipment health and respond directly to alarms and alerts. The customer using Shoreline AI’s APM Virtual Assistant application can use natural language to get insights into asset performance. The following screenshots show user interaction with the APM Virtual Assistant.

Figure 7: Screenshots of user interactions from Shoreline AI’s APM Virtual Assistant.

The following are some of the frequently asked queries being used by the customer’s plant managers and reliability engineers while interacting with the Shoreline AI APM Virtual Assistant:

• “Show Assets that have exceeded client defined thresholds in the last 1-month e.g. increasing Vibration-Trend AND/OR Temp-X AND/OR Speed-Y”
• “Analyze performance of a particular make/model of a pump or compressor operating across all client sites?”
• “MTTR (Mean-Time-To-Repair) and MTTA (Mean-Time-To-Acknowledge) data to understand if the responses are happening fast enough. Maintenance Management information.”
• “Which are the three most high-risk assets to review for asset health?”
• “How many unplanned downtime events occurred in the last 90 days and what was the total duration”
• “Show me repair history for a specific asset.” Here, the Virtual Assistant returns results from the plant Computerized Maintenance Management System (CMMS) solution and file system where unstructured data are stored.
• “Elaborate details of a specific alarm type.” Here, the Virtual Assistant not only looks at all structured and unstructured data within the organization, but also looks at best practices for possible fixes and probable causes from the carefully curated information on the internet.
• “Show monthly leakage detection trend for all my Slop Tanks”
• “List all Slop Tanks being Monitored”
• “Give me the status of leak detection for [a specific] Slop Tank.”
• “Which tank had the longest leak duration”

Business benefits of the APM Virtual Assistant for the customer

The business benefits of the APM Virtual Assistant are as follows:

• Natural Language Interaction (NLI): The APM Virtual Assistant enables NLI with the industrial APM software using natural language commands and queries. This makes the software more intuitive and easier to use, especially for users who are not familiar with traditional data analysis tools. For example, users can ask the questions about the performance of their assets, or they could instruct the software to perform specific tasks, such as generating complex reports and charts.
• Root-Cause Analysis: Generative AI is also being used to identify the root causes of equipment failures by analyzing vast amounts of data and generating hypotheses. It helps users address underlying issues and prevent similar failures from recurring. For example, it is used to analyze data from multiple sources, such as sensor data, maintenance records, and operator logs, to identify the factors that contributed to an equipment failure. Then, this information could be used to implement corrective actions to help prevent similar failures from occurring in the future.
• Optimized Maintenance Scheduling: Generative AI optimizes maintenance scheduling by strategically prioritizing tasks based on their predicted likelihood of failure and their potential impact on operations. This approach makes sure that maintenance resources are deployed more efficiently, significantly reducing disruptions to production processes and optimizing the overall performance.

Responsible and ethical generative AI practices
Adhering to the fundamental principles for responsible and ethical generative AI, Shoreline AI’s uses Amazon Bedrock for its APM Virtual Assistant to provide the following:

• Security: Only authorized users have access to the generative AI bot.
• Data Privacy: There is no company sensitive or confidential data being sent outside of Shoreline AI’s AWS environment.
• Traceability: Every piece of information provided by the APM Virtual Assistant is backed by verifiable data sources. Whenever the assistant pulls data or insights, it can provide references or links to the original sources. This promotes transparency and traceability, allowing users to verify the authenticity and accuracy of the information accessed.
• Observability: Logging enables the comprehensive monitoring and auditing of the solution’s behavior.
• Implementing Guardrails: Using Amazon Bedrock tools, the application is configured to filter harmful content, detect and deny unauthorized access, and deny access to sensitive information.

By implementing these protocols, Shoreline AI not only boosts the integrity and reliability of its virtual assistant, but also fosters trust and confidence among its users, aligning with global standards for ethical AI practices.

Call to action

For energy companies looking to unlock the full potential of their industrial assets and reduce their environmental impact, we encourage you to connect with the Shoreline AI team. The cloud-based APM and Emissions Leak Detection solutions, powered by predictive analytics and generative AI, have proven transformative for leading energy companies. The natural language interface allows plant managers and technicians to quickly access insights and make data-driven decisions, without the need for specialized data analysis skills.

To learn more about how Shoreline AI can help your organization optimize performance, drive efficiency, and enhance sustainability, visit the Shoreline AI website or check out APM 2.0 solution listing on AWS Marketplace.