The Beginning
In the exciting realm of AI, our journey unfolds — a testament to the triumphs and setbacks that shaped our path. At the heart of our story lies a grand ambition born from the vision of a pioneering oil and gas company. We set out to forecast the flowing bottom hole pressure at the depths of oil wells, harnessing the predictive power of AI. However, the issue of neglected maintenance emerged as a crucial factor affecting our success, which ultimately meant optimizing production and shaving millions off operational costs.
Challenge:
An oil-producing company relied on short-lived, physical gauges for measuring the crucial bottom-hole pressure (BHP) in its oil wells. BHP directly dictates the productivity index of the well and helps optimize the production. However, frequent failures due to the poor run life of these sensors in high-vibration conditions led to the unavailability of critical data and millions of dollars in annual expenses. This situation highlights the consequences of neglected maintenance. In addition to the hefty equipment and installation costs, expensive workovers and downtime in oil wells during installation further compounded the problem.
The Ambitious Goal
The company was actively seeking alternative BHP measurement solutions that led to the inception of an innovative FBHP prediction project aiming to revolutionize the process by replacing physical gauges with ultra-long lifespan, AI-powered digital sensors for accurate BHP prediction and proactive well workovers. This ambitious project had the potential to reduce costs, minimize disruptions, and optimize production.
Solving the Problem
Machine Learning Algorithm to Predict the Pressure!
Our data science team seamlessly contributed to the innovative ambition by collaborating with domain experts, bringing together a wealth of insights, which played a pivotal role in crafting an AI model, enabling precise predictions of bottom-hole pressure within acceptance criteria of ±40 PSI and made the real-time readings available to the team 24*7.
The Deployed Model
The trained model deployed in 102 wells post-training the model on 6 months’ data contributed to an 80% reduction in the annual installation of downhole sensors. However, the installation of physical gauges was required in only 10% of wells for continuous retraining of the model.
Failure
A retreat from success in just A few months down the line.
The predictions of the model started downgrading as there was a need to capture and clean the data from a few operational gauges to maintain the sanity of the model – a continuous need for model maintenance (Retraining)
What went wrong:
Failure to cultivate internal AI capabilities of the company
Relying heavily on outsourcing for AI model development and maintenance creates a gap often leading to missed opportunities to fixes, updates. This reliance on external resources is a form of neglected maintenance, which ultimately hampers the ability to become self-sufficient in future projects and increases costs.
Failure to maintain a model
Ultimately, the lack of internal expertise and dedicated resources for upkeep caused the model to become outdated and inaccurate, leading to the unfortunate regression from initial success.This situation underscores the consequences of neglected maintenance, as the absence of routine checks and updates can result in diminished performance.
Lesson Learnt
Internal Capability development:
Be proactive about the entire lifecycle of AI projects. Developing a solid foundation internally can prevent the pitfalls of neglected maintenance, which otherwise becomes cost-intensive when outsourced.
Continuous Improvements:
AI models need constant refresh and updates to keep them thriving. Ignoring the necessity for ongoing maintenance can lead to failures down the line.
Emerging Patterns:
Regular updates infused with fresh, diverse data are the secret sauce that allows the model to improve and predict better than its predecessor, combating the risks associated with neglected maintenance.
Never Ignore AI:
Delaying or ignoring necessary updates is a stumbling block—the hitch that could lead to the demise of a successful model. Neglected maintenance should never be overlooked.
Conclusion
A momentary setback is not a defeat but a stepping stone towards refining and fortifying our approach. In the dynamic world of AI, challenges are not roadblocks but rather a crucial learning opportunity; they are invitations to evolve, adapt, and emerge stronger than before. Recognizing and addressing neglected maintenance is essential to ensure the long-term success of AI projects.
As we move forward, it is essential to cultivate internal capabilities, maintain our models rigorously, and embrace continuous improvement. By doing so, we not only prevent regression but also unlock the full potential of AI to drive innovation and efficiency. The future of AI is bright, but it demands diligence, commitment, and an unwavering focus on sustaining progress.As we move forward, cultivating internal capabilities, rigorously maintaining AI models, and embracing continuous improvement are critical. These steps not only prevent setbacks but also unlock AI’s full potential to drive innovation and efficiency. While the future of AI is promising, it requires diligence, commitment, and an unwavering focus on sustained progress. To explore other key factors in building successful AI projects, visit our blog: The Success of AI Projects: Three Main Pillars.
We’re excited to announce that another blog in our “Failure Story” series is coming soon. Stay tuned for more insights and lessons learned from real-world AI projects!
We encourage you to share your thoughts and experiences regarding AI failures or any concerns you may have about AI adoption. If you have questions or would like to learn more, please feel free to reach out to us at support@emlylabs.com. We’re here to help!