In the oil and natural gas industry, maintaining the continuous operation of offshore rotating equipment is critical for production efficiency and cost management. Traditional maintenance methods—reactive and time-based—have proven inadequate for the demands of offshore environments. Reactive maintenance responds only after equipment breakdowns, while time-based maintenance follows fixed schedules regardless of equipment condition, both leading to increased downtime and higher costs. In contrast, predictive maintenance, powered by artificial intelligence (AI) and machine learning (ML), offers a proactive approach by predicting equipment failures before they occur, thereby enabling scheduled maintenance. This shifts maintenance strategies from reactive to predictive, enhancing equipment reliability, and reducing operational disruptions.
Introduction to Predictive Maintenance
Predictive maintenance leverages AI and ML to analyze data from equipment sensors, identifying patterns that precede failures. This proactive approach allows for maintenance to be scheduled just in time, avoiding unnecessary downtime and extending the lifespan of equipment. The transition from traditional maintenance methods to predictive maintenance represents a significant shift in how offshore equipment is managed, promising substantial improvements in efficiency and cost savings. By continuously monitoring equipment health and performance, AI/ML-based predictive maintenance provides early warnings about potential issues, enabling timely interventions that can prevent costly breakdowns.
In implementing predictive maintenance, the focus shifts from routine, often unnecessary checks and servicing to data-driven decisions. This not only minimizes operational disruptions but also optimizes maintenance resources and extends the service life of critical offshore equipment. By leveraging AI and ML, companies can analyze vast amounts of sensor data in real-time, identifying subtle signs of wear or malfunction that human operators might miss. This predictive approach ensures that maintenance is carried out only when needed, based on the actual condition of the equipment, rather than on predetermined schedules.
Murphy’s Predictive Maintenance Project Overview
Murphy undertook an ambitious project to implement AI/ML-based predictive maintenance on deepwater platforms in the Gulf of Mexico. This initiative aimed at boosting operational reliability and providing early warnings for equipment failures. Spanning 24 months, the project focused on production-critical rotating equipment like turbines, compressors, and pumps. To achieve its goals, Murphy adopted a transformative methodology that included data integration, predictive model deployment, and continuous insights refinement for the maintenance teams. The initiative was driven by a need to maximize equipment uptime and enhance overall operational efficiency through cutting-edge technology.
The project began with a comprehensive assessment of the existing maintenance practices and the identification of key areas for improvement. By integrating advanced AI and ML techniques, Murphy aimed to shift from a reactive maintenance model to a predictive one, thereby reducing unplanned downtime and improving equipment reliability. The scope of the project included not only the implementation of predictive maintenance tools but also the establishment of a robust data infrastructure to support ongoing analysis and decision-making. This strategic approach was designed to ensure that the benefits of predictive maintenance could be sustained over the long term.
Methodology and Implementation
The project was executed in distinct phases, each contributing critically to the overall aim of enhancing operational efficiency and reliability.
Data Capture and Transfer
The process begins with sensor data capture from offshore equipment. Initially stored in an offshore historian, this data is securely held and readily accessible for initial analysis. Systematic transfer to an onshore historian follows, facilitating long-term storage and comprehensive data review. This process ensures informed decision-making and operational efficiency. By leveraging this approach, Murphy can maintain a high level of data integrity while ensuring that the insights generated are based on accurate and up-to-date information. The seamless transfer of data between offshore and onshore locations is a critical component of the project, enabling continuous monitoring and timely interventions.
Predictive Maintenance Workflow
Before the predictive maintenance project, once data flowed from the sensor to the historian, specific data from natural-gas compressors and main power-generation turbines were sent via an Open Platform Communications (OPC) server to the equipment service partner’s cloud. This service partner set alerts based on predefined thresholds and notified the company for necessary maintenance actions. The workflow ensured that equipment issues were identified and addressed promptly, minimizing the risk of unplanned downtime. By integrating predictive maintenance into the existing workflow, Murphy was able to enhance the overall efficiency of its operations and improve the reliability of its offshore equipment.
Data Integration and Automation
The integration phase involved significant data handling and transformation. Initially, data scraping was done manually from the service partner’s cloud. To streamline this, a dedicated data pipeline periodically transferred the scraped data. Additionally, the time-series data from sensors was combined with event-based data, including equipment history from the Computerized Maintenance Management System (CMMS) and daily progress reports. Initially performed manually, these tasks were later automated with a Representational State Transfer Application Programming Interface (REST API), ensuring continuous and seamless data integration. This level of automation was essential for maintaining the efficiency and accuracy of the predictive maintenance system, enabling Murphy to detect and address potential issues before they escalate into major problems.
Development and Deployment of Predictive Models
Leveraging this enriched data pool, predictive models were developed and deployed in the cloud. These models generated anomaly alerts reviewed by cross-functional teams, including members from the AI/ML service partner and Murphy’s engineering, reliability, maintenance, and operations teams. Initial alerts indicated the presence of anomalies but required refinement to reduce false positives and enhance specific failure detection. The collaboration between the various teams was crucial for ensuring that the predictive models were accurate and effective, with continuous feedback and improvements being made based on real-world observations and data. This iterative approach allowed Murphy to gradually refine the predictive models, increasing their reliability and effectiveness over time.
Seamless CMMS Integration
To ensure timely maintenance execution, true positive alerts were used to manually create notifications in the CMMS initially. To improve efficiency, a REST API was later developed to automate notification creation, ensuring Murphy’s offshore maintenance teams received alerts for potential issues proactively. This integration enabled a more streamlined and efficient maintenance process, reducing the time and effort required to address equipment issues. By automating the notification process, Murphy was able to ensure that maintenance teams were always up to date with the latest information, allowing them to prioritize and address the most critical issues promptly.
Observations and Trend Analysis
During the project, 46 predictive models were deployed across two Gulf of Mexico platforms. Each platform had unique models tailored to its equipment. Initial models started generating alerts months after deployment. The alert generation trend showcased the progression of alerts over time, highlighting the effectiveness of the predictive models in identifying potential issues. Platform-specific variations in alerts demonstrated differences in equipment monitoring and anomaly detection effectiveness. This variability underscored the importance of customizing predictive models to suit the specific needs and operating conditions of each platform. By tailoring the models to the unique characteristics of each platform, Murphy was able to achieve more accurate and reliable predictions.
Alert Disposition
The reviewed alerts were categorized by a cross-functional team. The incidence of false positives initially surged but decreased following model retraining. This ongoing refinement underscored the critical role of continuous learning and adaptation in predictive maintenance model efficacy. By continuously improving the models based on real-world data and feedback, Murphy was able to enhance the accuracy and reliability of the predictive maintenance system. This iterative approach ensures that the system remains effective over time, adapting to changes in equipment behavior and operating conditions.
Discussion on Project Planning and Execution
Project Planning and Management
A pilot project involving AI/ML-based predictive maintenance was spearheaded to validate the efficacy of predictive models. The pilot’s success established high expectations but also highlighted the necessity for deeper project understanding before full-scale implementation. Key challenges included initial misalignments and delays due to the inadequate assessment of project prerequisites. Future projects must ensure a comprehensive understanding of prerequisites, including detailed planning and stakeholder engagement to mitigate early-stage challenges. By addressing these challenges proactively, Murphy can ensure smoother implementation and greater success in future predictive maintenance initiatives.
Data Prerequisites: Completeness and Quality
The project’s effectiveness was significantly impacted by the availability and quality of data. Delays in deploying the first predictive model stemmed from insufficient requisite data, necessitating a comprehensive data readiness check during the project startup phase. This involved assessing data sources, quality, and completeness. During the project, it was discovered that progress reports and CMMS event data were incomplete and lacking quality. Ensuring robust data integrity upfront would have streamlined the model development process. By prioritizing data quality and completeness, Murphy can enhance the effectiveness and accuracy of its predictive maintenance models, leading to better overall outcomes.
Recommendations for Future Projects
Enhancing Data Quality
Improving the quality of event data in the CMMS is essential. Ensuring data accuracy and completeness impacts predictive models’ performance. Instituting regular data audits and cleansing can maintain data integrity. By implementing rigorous data management practices, Murphy can ensure that its predictive maintenance system is built on a solid foundation of high-quality data, enabling more accurate and reliable predictions.
Instrumentation Adequacy
Precise instrumentation on critical equipment is crucial. Investing in modern sensors and data acquisition systems is recommended. A thorough cost/benefit analysis can justify these investments, ensuring resource allocation effectiveness. By equipping offshore platforms with the latest sensors and monitoring technology, Murphy can enhance the accuracy and reliability of its predictive maintenance system, leading to better outcomes and reduced downtime.
Domain Knowledge Development
Building in-house knowledge of rotating equipment is invaluable. Specialized expertise will enable more targeted and effective predictive models. Implementing training programs and knowledge-sharing initiatives will foster this expertise within the organization. By investing in the development of in-house expertise, Murphy can enhance its ability to maintain and optimize its offshore equipment, leading to greater reliability and efficiency.
Technological Advancements in AI/ML
Regularly revisiting AI/ML solutions is vital to staying abreast of technological advancements. Updating methodologies and tools will refine model accuracy and predictive capabilities. By continuously evolving its predictive maintenance system, Murphy can ensure that it remains at the cutting edge of technology, delivering the best possible outcomes for its offshore operations.
Conclusion
Murphy’s implementation of AI/ML-based predictive maintenance showcased the transformative potential of these technologies in the offshore oil and natural gas industry. By shifting from traditional methods to proactive maintenance strategies, predictive maintenance enhanced equipment reliability and operational efficiency. The project’s phased methodology demonstrated significant lessons in data integration, domain knowledge, and model refinement. Moving forward, ensuring data quality, adequate instrumentation, expertise development, and technological adaptability will elevate future predictive maintenance projects’ success rates. Through these focused improvements, the industry can fully harness the benefits of AI/ML-based predictive maintenance, achieving greater operational reliability and cost efficiency.
