Saltwater disposal (SWD) well operations are energy-intensive and typically lack continuous performance visibility.
This article presents a results-driven case study from an ongoing collaboration between a midstream oil and gas company and Neuralix Inc., in which artificial intelligence (AI) and first principles-based time series analytics were used to deliver significant operational cost reductions.
Neuralix's AI system delivered up to a 40% electricity savings across a subset of injection sites in Phase 1, with projected annualized improvements exceeding 40% once scaled.
This work demonstrates how interpretability, domain specificity, and first principles thinking can unlock actionable value from complex supervisory control and data acquisition (SCADA) environments.
Introduction
SWD is a cornerstone of produced-water management in oil and gas operations. However, it comes with significant power costs due to high‑pressure injection pumping. As a midstream operator managing several sites across Oklahoma and Texas, the client operator aimed to reduce electricity usage while improving operational oversight. Neuralix achieved this through the deployment of an AI-powered key performance indicator (KPI)-monitoring and optimization system.
The collaboration sought to
- Reduce kWh/bbl and cost/bbl.
- Identify underperforming pump configurations.
- Deliver transparent, actionable insights to engineers and field teams.
First Principles Approach
Neuralix’s approach at solving complex operational challenges is deeply rooted in first principles thinking. Instead of relying on opaque, "black box" AI models, Neuralix breaks each challenge down into its fundamental components—physics, chemistry, and operational constraints—and builds solutions from the ground up.
In the case study project, this meant
- Deconstructing pump energy inefficiencies to their core thermodynamic and hydraulic causes.
- Designing interpretable analytics around kWh/bbl, $/bbl, and flow rate as governing KPIs.
This method allowed the solution developer to pinpoint why certain pumps consume more power per bbl and what operational conditions (e.g., abrubt changes, poor filter quality, suboptimal frequencies) are driving that behavior. It is particularly critical in SWD disposal operations where SCADA data is noisy, multivariate, and often lacks labels.
Technical Implementation, Data Ingestion, and Structuring
Neuralix’s proprietary Data Lifecycle Templatization (DLT) system standardized ingestion of time series data from diverse SCADA systems.
Core parameters included:
- Motor frequency (Hz)
- Flow rate (B/D)
- Voltage and amperage
- Pressure readings
- kWh pricing integration
Dashboard and KPI Design
A production-grade, secure dashboard was created to visualize site-level and group-level KPIs in real time. Features included
- Site-wise and group-wise KPI summaries.
- Historical performance tracking.
- Customizable date-range comparisons.
- $/bbl and kWh/bbl threshold alerts.
Phase 1 Outcomes
- Energy reduction of 30%
- Sites achieved 14% reductions in $/bbl
- Estimated improvement in average operating efficiency per pump of 14%
Phase 2 Progress
- Doubled number of sites tracked
- Estimated 2-month energy efficiency improvement of 12%
- Annualized projected optimization across all sites of more than 40%
Case Insight: Pump Frequency Optimization
Through observation alone, we can study past operations to recommend operational improvements. For example:
- At one site, shifting from high-frequency, short-duration pumping to continuous low-frequency operation—while maintaining the same mid-capacity daily volume—reduced energy consumption, representing a 12% improvement in efficiency.
- At another site with a different pump type, increasing the operating frequency represents a 17% improvement in efficiency.
This analysis highlights the importance of pump-specific optimizations. Recommendations will vary depending on the health state of the system, the type of pump, and operational constraints.
Broader Application and Modularity
Neuralix’s work is part of a broader suite of use cases where first-principles AI has proven transformative such as deepwater electrical submersible pump failure warning triggered by silt buildup patterns; leak detection and localization in pipeline systems; and fault detection in genomic substrate automation
In all cases, the technology delivered high-return-on-investment solutions by iteratively applying modular DLT templates tailored to each client’s operations, guided by a fundamental understanding of the physical systems involved.
Conclusion
This case demonstrates that pragmatic, first-principles-driven AI can deliver real-world impact in water infrastructure operations. By combining domain-specific interpretability, reliable KPI frameworks, and causal analytics, the solution developer enabled the midstream operator to not only cut costs but better understand the behaviors behind those costs.
While cutting electricity costs represents an ideal entry point to capture immediate savings, when pumps are operating near full capacity, slowing pumps to save power can reduce throughput and overall revenue. In these situations, further efficiency gains must come from optimizing maintenance strategy instead.
Neuralix’s models predict how each pump’s performance degrades after a service and identify when efficiency has declined enough for the next maintenance to yield a net positive return.
Scheduling maintenance at the right interval directly affects revenue. Wait too long between overhauls and the pump spends too much time underperforming; service too often and you lose productivity to excessive downtime. Essentially, performance peaks right after maintenance and then declines. The optimal time to intervene is when the boost from servicing just offsets its cost. Traditional reactive maintenance often misses this window, intervening too late.
By extrapolating the degradation curve with the use of a digital twin, it is possible to predict when impending performance loss will outweigh maintenance costs. Initial proof-of-concept analyses show that scheduling maintenance at this optimal point (for example, performing an acid treatment right before a steep decline) can boost average daily revenue by around 5% compared to a typical schedule.
Even a 1% revenue uptick is about $40,000 per year for one SWD site, and across 250 sites it’s nearly $10 million in additional annual revenue.

Annorah Lewis is a product and operations leader at Neuralix Inc., where she helps develop AI-powered solutions for energy and industrial applications. She holds a BS in computer science from Virginia Tech and has completed the Harvard Business School Online program. Her work focuses on aligning advanced technologies with industry needs, supporting efforts such as predictive maintenance, emissions reduction, and asset optimization. At Neuralix, she works across teams to shape product strategy, guide implementation, and translate customer challenges into practical AI-driven solutions.

Audrey Der is a lead data scientist at Neuralix Inc., where she focuses on time series data mining and machine learning. Her work centers on human-in-the-loop frameworks for anomaly detection, pattern recognition, and event classification in operational data. She also leads the development of the company’s core machine learning library and oversees its intern program, designing tailored interview processes and project scopes.
Der holds BS and PhD degrees in computer science from the University of California, Riverside, under the guidance of Eamonn Keogh, known for pioneering the Matrix Profile technique. Her doctoral work includes the development of PUPAE, an explainable anomaly classification system, and PRCIS, an interpretable distance measure for long time series. She previously interned at Visa, where she contributed to synthetic time series generation methods and evaluated generative models for classification performance.

Ryan Mercer is a lead data scientist at Neuralix Inc., where he develops AI-based solutions to improve operational efficiency and reliability in the energy sector. His work focuses on anomaly detection, predictive maintenance, and pattern recognition to support proactive asset management. He holds a BS in electrical engineering from UCLA and a PhD in computer science from the University of California, Riverside. Earlier roles include research internships at Visa, Toyota, and Form Energy and work as a test engineer at SanDisk. At Neuralix, he has led projects such as early failure detection in saltwater disposal systems, HPump energy optimization, and methane leak-localization modeling.

Vikram Jayaram, SPE, is the founder and CEO of Neuralix Inc., an AI startup advancing digital transformation in manufacturing and oil and gas. He has spent 2 decades applying artificial intelligence and machine learning to industrial systems, with a focus on upstream operations. He previously served as head of R&D and data science at Pioneer Natural Resources, leading technology development supporting $7 billion in annual operations across drilling, completions, production, and HSE. His experience includes roles at Sabre Corp., Global Geophysical Services, and the University of Oklahoma. Jayaram holds PhD and MS degrees in electrical engineering and completed postdoctoral work at the University of Texas M.D. Anderson Cancer Center. A NASA Doctoral Fellow, he has authored more than 100 peer-reviewed papers, journals, and patents.