By: Dauda Lawal – Web Developer
In the oil and gas industry, equipment reliability directly influences production continuity, operating costs, personnel safety and environmental performance. Natural gas compression facilities depend on complex mechanical systems that must operate consistently under demanding conditions. When a critical compressor fails unexpectedly, the consequences can extend beyond repair costs to include lost throughput, contractual exposure, emergency maintenance and potential safety incidents.
For operators managing equipment across remote locations, maintaining reliable visibility into asset health becomes even more challenging. Sensor data may remain isolated within local control systems, maintenance teams may rely on calendar-based schedules, and unstable network connectivity can prevent continuous monitoring from centralized platforms.
Chevan, an energy company operating natural gas compression and gathering assets across remote sites in Nigeria, faced these challenges across a fleet of 46 compressors distributed across 12 locations.
To address these operational limitations, Chevan partnered with Arthurite Integrated to implement an AWS-powered Predictive Maintenance and Asset Reliability Platform. The solution combines industrial IoT, edge computing, machine learning and automated maintenance workflows to help identify developing equipment faults, estimate remaining useful life and support more informed maintenance decisions.
By processing equipment data locally and synchronizing relevant insights with AWS, the platform is designed to maintain monitoring capabilities even when remote connectivity becomes unavailable.
This article explores the operational challenges behind the engagement, the AWS architecture implemented, the engineering capabilities supporting the platform and the business outcomes the solution aims to achieve.
The Challenge: Moving Beyond Calendar-Based Compressor Maintenance
Chevan operates a fleet of 46 natural gas compressors, including reciprocating and centrifugal units, across 12 remote compression and gathering sites.
The fleet is instrumented with approximately 1,750 sensor tags, covering important operating parameters such as vibration, temperature, pressure, lubrication, speed and load.
Despite having access to equipment data, the organization faced challenges turning those measurements into timely maintenance decisions.
1. Reactive Maintenance and Unplanned Compressor Shutdowns
Before the engagement, maintenance activities relied primarily on calendar-based schedules and operating-hour intervals rather than continuous equipment condition monitoring.
This approach created two problems.
First, healthy compressors could undergo unnecessary maintenance simply because they had reached a scheduled service interval. Second, developing mechanical faults could remain undetected between inspections, allowing equipment conditions to deteriorate until a failure occurred.
The fleet experienced approximately 11 unplanned shutdowns annually, with each shutdown taking an average of 5.5 days to restore.
That translated into approximately 60 compressor-days of lost availability every year.
Considering that each compressor-day of throughput represents approximately ₦5.1 million, these interruptions created a significant financial burden.
2. Unreliable Connectivity Across Remote Sites
Chevan’s remote facilities depend on satellite and cellular connectivity, with average link availability of approximately 71%.
Network interruptions can last between three and five days.
This created a major limitation for conventional cloud-only monitoring architectures. A system that continuously transmits raw sensor data to the cloud would lose visibility whenever a remote site’s connection became unavailable.
For industrial operations, this is particularly problematic because equipment monitoring should remain available even when external connectivity fails.
Chevan therefore required an architecture capable of processing sensor information locally, maintaining alarm functionality offline and synchronizing historical data after connectivity returned.
3. Fragmented Sensor Data and Limited Fleet-Wide Visibility
Equipment data remained distributed across local SCADA systems, PLCs, RTUs and site historians.
Reliability engineers had to review information from different sources and investigate equipment conditions after an alarm or failure occurred.
Without a standardized asset model, comparing compressor performance across different sites and equipment types also became more difficult.
The organization needed a centralized platform capable of organizing equipment information consistently across its fleet while preserving the existing industrial control infrastructure.
4. Delayed Fault Identification and Maintenance Dispatch
Existing equipment alarms were communicated primarily through phone calls and email.
Although these notifications helped maintenance personnel respond to known issues, they did not provide sufficient predictive information about developing faults.
Maintenance teams often lacked an early indication of the probable failure mode, the severity of the condition or the time remaining before a potential failure.
This limited their ability to prioritize work orders, prepare replacement parts and coordinate field technicians before equipment downtime occurred.
The Financial Impact of Reactive Maintenance
The workbook estimates the annual financial impact of Chevan’s reactive maintenance approach at approximately ₦527 million.
| Cost category | Estimated annual impact |
|---|---|
| Lost throughput from unplanned downtime | ₦310 million |
| Emergency repair premiums and expedited parts | ₦85 million |
| Unnecessary calendar-based maintenance | ₦62 million |
| Safety and environmental incident exposure | ₦48 million |
| Manual rounds and data logging | ₦22 million |
| Total estimated annual impact | ₦527 million |
Beyond these direct costs, continued reliance on reactive maintenance creates long-term risks involving asset deterioration, gas delivery commitments, operational continuity and the safety of personnel working around high-pressure compression equipment.
Chevan needed a solution that could identify developing faults earlier, maintain visibility during connectivity interruptions and help maintenance teams make decisions based on actual equipment condition.
The AWS Solution: An Edge-to-Cloud Predictive Maintenance Platform
Arthurite Integrated implemented an industrial IoT and predictive analytics architecture built around AWS IoT Greengrass, AWS IoT SiteWise and Amazon SageMaker.
The solution covers 46 natural gas compressors across 12 remote sites, with approximately 1,750 monitored sensor tags.
Rather than transmitting every raw sensor measurement to the cloud, the platform processes high-frequency vibration information at the edge. This reduces bandwidth consumption while allowing important equipment-health indicators to remain available locally.
AWS IoT Greengrass V2 runs on ruggedized industrial gateways at the remote sites. SiteWise Edge components connect to existing PLC, RTU and SCADA systems through read-only OPC UA and Modbus TCP interfaces.
The gateways calculate vibration features, apply local alarm rules, run compact anomaly-detection models and buffer data during connectivity interruptions.
At the cloud layer, AWS IoT Core authenticates devices and routes telemetry into AWS IoT SiteWise. Amazon SageMaker provides predictive analytics capabilities, including anomaly detection, failure-mode classification and remaining-useful-life estimation.
Amazon EventBridge, AWS Lambda and Amazon SNS coordinate alerting, while API Gateway integrates the platform with Chevan’s existing computerized maintenance management system (CMMS).
Amazon Managed Grafana provides fleet-wide equipment visibility, compressor-level diagnostics and maintenance planning dashboards.
The architecture connects remote industrial assets, local processing, cloud-based analytics and existing maintenance operations into one coordinated platform.
Five Key Engineering Capabilities Behind the AWS Solution
1. Industrial Edge Computing with AWS IoT Greengrass V2
The first major engineering capability is the ability to process equipment data locally without depending on continuous cloud connectivity.
Each of the 12 remote sites has a ruggedized industrial gateway running AWS IoT Greengrass V2 with SiteWise Edge components.
The gateways connect to existing industrial systems through read-only OPC UA and Modbus TCP interfaces.
They collect measurements from approximately 1,750 sensor tags, including:
- Vibration measurements.
- Bearing and discharge temperatures.
- Lubrication oil temperature and pressure.
- Stage pressures and pressure ratios.
- Compressor speed and operating load.
Raw vibration signals can be sampled at frequencies between 1 kHz and 10 kHz. Processing these signals locally allows the platform to calculate useful condition indicators without continuously transmitting the entire raw dataset.
The edge components extract vibration features such as:
- Root Mean Square (RMS).
- Peak amplitude.
- Kurtosis.
- Fast Fourier Transform (FFT) frequency bands.
These features help identify changes in vibration patterns that may indicate bearing wear, mechanical imbalance or other developing equipment problems.
The gateways also calculate one-minute aggregates for temperature and pressure measurements.
This edge-processing approach reduces the amount of data transmitted to the cloud by approximately 90%.
Offline monitoring and data resilience
Connectivity interruptions do not automatically stop the monitoring process.
AWS IoT Greengrass Stream Manager buffers processed sensor information locally for up to seven days. When connectivity returns, the gateway exports buffered records in chronological order.
Local OEM alarm limits and compact anomaly-detection models continue operating while the site is disconnected.
The target is to generate local alarms within 10 seconds of an OEM limit exceedance, even when the cloud connection is unavailable.
This architecture allows Chevan to maintain equipment monitoring capabilities across remote locations where network reliability cannot be guaranteed.
2. Centralized Asset Modeling with AWS IoT SiteWise
Managing 46 compressors across 12 sites requires a consistent way to organize equipment information and understand relationships between individual assets.
Arthurite uses AWS IoT SiteWise to establish a structured industrial asset hierarchy.
The platform organizes assets using the following model:
Fleet → Site → Compressor Train → Compressor → Subsystem
Each compressor is represented through reusable asset templates based on its equipment type.
These templates standardize how sensor measurements, operating conditions and computed performance indicators are organized across the fleet.
AWS IoT SiteWise receives authenticated telemetry through AWS IoT Core and provides a consistent foundation for equipment monitoring and analytics.
The platform also supports computed metrics that help reliability engineers interpret operating conditions rather than reviewing disconnected sensor readings.
This approach offers several advantages:
- Centralized visibility across all 12 sites.
- Consistent modeling of reciprocating and centrifugal compressors.
- Easier comparison of equipment performance.
- Standardized onboarding of additional compressors.
- Better integration between sensor information and predictive analytics.
Rather than building a separate custom asset hierarchy in a database, the architecture uses AWS IoT SiteWise to manage industrial asset relationships and associated measurements.
3. Predictive Analytics and Remaining Useful Life Estimation with Amazon SageMaker
The central intelligence layer uses Amazon SageMaker to identify equipment abnormalities and estimate potential failure conditions.
The platform uses historical historian data, maintenance records and failure information to develop models suited to Chevan’s compressor fleet.
The documented training dataset covers approximately 18 to 24 months of historical operating information.
The machine-learning approach combines three important capabilities.
Multivariate anomaly detection
The platform uses custom autoencoder and isolation-forest models to identify unusual operating patterns across multiple sensor measurements.
Instead of evaluating each sensor independently, the models consider relationships between vibration, pressure, temperature, lubrication and operating load.
This helps identify abnormal equipment behaviour that may not trigger conventional fixed-threshold alarms.
Failure-mode classification
When the system identifies an abnormal condition, supervised models help classify the probable failure category.
The documented failure categories include:
- Bearing-related faults.
- Valve and ring wear.
- Lubrication problems.
- Cooling-system failures.
The solution uses historical maintenance and failure records, alongside engineering input from Chevan’s reliability team, to improve the relevance of its predictions.
Remaining Useful Life estimation
Remaining Useful Life (RUL) estimation helps maintenance teams understand how much operating time may remain before an asset reaches a predicted failure condition.
Rather than simply reporting that a compressor is operating outside its normal range, the platform aims to provide an estimate that supports maintenance prioritization.
The documented technical target is an RUL estimate within ±20% at seven days to failure.
Amazon SageMaker Pipelines supports quarterly model retraining, while approved model packages can be deployed to the site gateways through AWS IoT Greengrass.
This allows the platform to combine centralized model development with local inference capabilities.
Amazon Lookout for Equipment was initially considered, but the documented design selected custom SageMaker models instead because of its stated end-of-support timeline and the need for customer-specific predictive models.
4. Automated Alerts and Maintenance Work Orders
Predictive analytics creates operational value when the results reach the people responsible for maintaining equipment.
Arthurite integrates the detection layer with Chevan’s existing maintenance workflow.
When the platform identifies a potential fault, it publishes an event to Amazon EventBridge.
AWS Lambda enriches the event with relevant information, including:
- Compressor and site identification.
- Probable failure mode.
- Equipment condition.
- Recommended maintenance action.
- Alert severity.
The system classifies notifications into four levels:
- Advisory.
- Warning.
- Critical.
- Trip.
Amazon Simple Notification Service (Amazon SNS) sends SMS and email notifications to designated maintenance personnel.
A 15-minute escalation process supports the handling of alerts that require acknowledgement.
For maintenance integration, Amazon API Gateway exposes the required REST API endpoint for communication with the existing CMMS.
The platform creates work orders containing the relevant asset information, detected condition and recommended action.
Amazon DynamoDB tracks alert states and workflow information.
The documented technical objective is to create a work order within two minutes of a qualifying detection.
Importantly, the solution remains strictly read-only toward industrial control systems. It does not automatically issue commands to compressors or modify operating conditions.
This separation preserves existing control-system authority while allowing the predictive maintenance platform to improve maintenance decisions.
5. Historical Analytics and Fleet Visibility with Amazon S3, Athena and Amazon Managed Grafana
A reliable predictive maintenance platform needs to preserve historical information for trend analysis, model improvement and maintenance planning.
Arthurite uses Amazon S3 as the central data lake for historical features, alerts, maintenance outcomes and machine-learning artifacts.
AWS Glue catalogs the stored information, while Amazon Athena supports SQL-based analysis without requiring a separate database infrastructure for every analytical workload.
For operational visualization, Amazon Managed Grafana provides dashboards tailored to reliability engineering.
The dashboards support several important monitoring activities:
Fleet Health: Displays the overall condition of compressors across the 12 remote sites.
Compressor Diagnostics: Provides equipment-level visibility into vibration, temperature, pressure and other operating indicators.
Maintenance Planning: Uses remaining-useful-life estimates to help rank equipment requiring attention.
Reliability KPIs: Tracks indicators such as Mean Time Between Failures (MTBF), Mean Time to Repair (MTTR) and unplanned downtime.
This architecture gives maintenance and reliability teams a more consistent view of equipment performance and helps them prioritize interventions based on condition rather than calendar schedules alone.
Secure Connectivity, Resilience and Operational Safety
Remote industrial environments introduce network and operational risks that must be addressed before deploying a cloud-connected monitoring platform.
Chevan’s architecture uses a hybrid connectivity design with AWS Direct Connect as the primary connection between the central operations hub and AWS.
AWS Site-to-Site VPN provides a secondary connection with automatic Border Gateway Protocol (BGP) failover.
If the primary Direct Connect path becomes unavailable, network routing can move traffic to the backup VPN connection.
At the individual sites, existing VSAT and cellular networks provide connectivity between the gateways and the operations hub. Secondary carriers are used where available.
The platform also incorporates several AWS security and operational controls.
Device authentication: AWS IoT Core authenticates gateways using individual X.509 certificates.
Encryption: TLS protects telemetry in transit, while AWS Key Management Service (AWS KMS) supports encryption at rest.
Access control: AWS IAM applies least-privilege permissions across the cloud environment.
Device security: AWS IoT Device Defender supports gateway and device security audits.
Credential protection: AWS Secrets Manager stores credentials used for external integrations, including CMMS connectivity.
Auditability: AWS CloudTrail records relevant account and API activity, while AWS Config supports configuration monitoring.
Operational monitoring: Amazon CloudWatch provides visibility into service health, processing performance and system alerts.
The architecture also uses multi-Availability Zone deployment for important cloud components, including SageMaker endpoints, DynamoDB and Amazon Managed Grafana.
Maintaining Monitoring During Network Outages
One of the most important characteristics of this solution is its ability to continue operating when remote connectivity fails.
If a site loses its external WAN connection, the Greengrass gateway continues applying local OEM alarm limits and its deployed anomaly model.
Stream Manager retains processed sensor information for up to seven days.
When connectivity is restored, buffered data is transmitted in chronological order to the cloud platform.
This design addresses a major weakness of cloud-only monitoring systems: losing visibility whenever a remote site becomes disconnected.
The documented target is to recover at least 99.5% of buffered data after reconnection.
The cloud platform also targets 99.9% operational availability through its multi-AZ architecture.
Expected Business Impact: What Chevan Aims to Achieve
The Chevan engagement establishes measurable objectives for evaluating improvements in compressor reliability and maintenance operations.
The following table distinguishes the documented baseline from the 12-month targets.
| Performance indicator | Current baseline | Proposed 12-month target |
|---|---|---|
| Annual unplanned compressor-days | Approximately 60 | 36 or fewer |
| Annual unplanned shutdowns | Approximately 11 | 6 or fewer |
| Condition-driven maintenance | 0% | 60% |
| Early-warning lead time | Reactive detection | At least 72 hours on detectable failure modes |
| Failure-mode classification | Not standardized | At least 80% top-1 accuracy |
| Alert precision | Not established | At least 85% |
| Data recovery after reconnection | Not established | At least 99.5% |
| Edge bandwidth reduction | Raw/high-volume data transmission | Approximately 90% |
| Alert-to-work-order creation | Manual phone and email process | Under 2 minutes |
These figures are documented project targets, not independently verified achieved results. Actual measurements should be added once the agreed evaluation period and supporting operational records are available.
The Financial Opportunity
The potential business value is significant because the platform addresses several major components of Chevan’s estimated ₦527 million annual reactive maintenance cost.
The 12-month objectives include:
- Reducing unplanned downtime by 40%, from approximately 60 to 36 or fewer compressor-days annually.
- Reducing unplanned shutdowns from approximately 11 to six or fewer per year.
- Moving condition-driven maintenance from 0% to 60%.
- Reducing emergency repair premiums and expedited parts costs by 30%.
- Reducing unnecessary calendar-based maintenance costs by 25%.
- Reducing manual rounds and data logging effort by 60%.
The solution is designed to help recover throughput, reduce unnecessary maintenance activity and identify equipment conditions before they develop into major failures.
Production Deployment and AWS Investment
The workbook identifies the Chevan solution as a production-grade deployment that is live in the customer’s AWS account, rather than a proof of concept.
The documented estimated AWS Annual Recurring Revenue (ARR) associated with the use case is $50,000 per year.
This is the annual AWS revenue estimate recorded for the workload, not a separately verified customer savings figure or a detailed monthly service-cost breakdown.
The architecture also reflects several cost and operational decisions.
Edge feature extraction reduces the volume of data transmitted from remote sites by approximately 90%. Local inference avoids depending on continuous cloud connectivity for essential monitoring. AWS managed services support centralized ingestion, asset modeling, storage, analytics and event processing.
These choices help balance industrial monitoring requirements with the realities of operating across remote sites with limited bandwidth.
AWS Energy and Utilities Competency Alignment
The Chevan engagement demonstrates several capabilities relevant to the AWS Energy and Utilities Competency, particularly within the Midstream, Core Energy Business Applications, Data Analytics and Insights, and Health, Safety and Environment areas.
The architecture brings together:
- Industrial IoT: AWS IoT Greengrass V2, AWS IoT Core and AWS IoT SiteWise for connecting and managing remote industrial assets.
- Edge Computing: Local feature extraction, offline processing and seven-day store-and-forward buffering.
- Predictive Maintenance: Amazon SageMaker models for anomaly detection, failure-mode classification and remaining-useful-life estimation.
- Hybrid Cloud Connectivity: AWS Direct Connect with Site-to-Site VPN failover.
- Maintenance Automation: EventBridge, Lambda, SNS and API Gateway integration with the existing CMMS.
- Data Analytics: Amazon S3, AWS Glue, Athena and Amazon Managed Grafana for historical analysis and operational visibility.
- Security and Governance: AWS KMS, IAM, IoT Device Defender, CloudTrail, AWS Config and CloudWatch.
The engagement demonstrates how AWS services can support an energy-specific reliability problem rather than simply moving industrial data into the cloud.
Conclusion: Building a More Predictive Approach to Compressor Reliability
For natural gas operators, compressor reliability is directly connected to production continuity, maintenance expenditure, contractual performance and operational safety.
Chevan’s engagement demonstrates how organizations can move beyond calendar-based maintenance and reactive equipment repairs by combining industrial IoT, edge computing and predictive machine learning.
Through AWS IoT Greengrass V2, AWS IoT SiteWise and Amazon SageMaker, Arthurite Integrated implemented a platform that connects remote compressor assets to a centralized predictive maintenance environment while maintaining local monitoring capabilities during network interruptions.
The solution also connects predictive insights to existing maintenance workflows, helping teams prioritize interventions and create work orders with more useful equipment information.
By keeping industrial control systems read-only and retaining existing OEM alarm protections, the platform supports better maintenance decisions without taking control away from established operational systems.
As the energy industry continues to adopt connected industrial technologies, predictive maintenance will play an increasingly important role in improving asset reliability and reducing the cost of unplanned downtime.
Arthurite Integrated remains committed to helping energy organizations design and implement secure, scalable AWS solutions that turn industrial data into actionable operational intelligence.
Looking to improve equipment reliability across your energy operations?
Arthurite Integrated can help your organization assess its industrial infrastructure, implement edge-to-cloud IoT architectures and develop predictive maintenance solutions powered by AWS.