By: Dauda Lawal – Web Developer

Health, Safety and Environment (HSE) management is fundamental to reliable oil and gas operations. Across processing plants, flow stations and storage terminals, workers routinely operate around rotating equipment, pressurized systems and flammable materials. In these environments, delayed hazard detection, missed personal protective equipment (PPE) violations and slow incident reporting can increase operational risk and expose organizations to significant environmental, financial and regulatory consequences.

For Ironridge Energy Operation, improving safety oversight presents an opportunity to move beyond periodic manual inspections toward continuous, technology-enabled monitoring.

Arthurite Integrated has proposed an AI-powered HSE Monitoring and Incident Reporting Platform built on Amazon Web Services (AWS) to help address this challenge. The proposed solution combines computer vision, real-time video streaming, automated alerting and generative AI to support earlier hazard detection, faster incident escalation and more efficient safety reporting.

The initiative is designed for an initial deployment across 64 cameras at six Ironridge facilities. Rather than replacing HSE professionals, the platform aims to give them better visibility into safety conditions, actionable alerts and evidence-backed reports so they can focus more time on prevention and response.

This article describes a proposed solution and its intended outcomes. Performance targets are objectives to be validated during implementation, not confirmed results from a completed deployment.

The Challenge: Why Traditional HSE Monitoring Needs a New Approach

Limited visibility across operational facilities

Ironridge’s existing safety monitoring process relies heavily on periodic walk-throughs by HSE officers and the review of camera footage after incidents occur. Although the organization has approximately 210 installed CCTV cameras, the footage is primarily used for security rather than continuous safety analytics.

The Statement of Work estimates that manual inspections cover less than 2% of working hours in the areas under consideration. It also records approximately 30 inspections per month across the facilities and an estimated PPE non-compliance rate of 11% in audit observations.

This approach creates gaps between inspections. A worker may enter a restricted area without the required permit, or an individual may work without the appropriate hard hat, high-visibility vest, gloves or safety boots, without the issue being detected immediately. Similarly, visible smoke, flames or liquid spills may remain unnoticed until a worker reports them or an inspection identifies the problem.

Delayed incident detection and reporting

The proposed engagement identifies several operational bottlenecks:

• Late hazard response: The baseline estimates an average of approximately 4.5 hours from leak or emission detection to response.
• Slow incident documentation: Preparing an incident report takes approximately six working days.
• Lengthy regulatory reporting: Preparing a regulatory report takes approximately three working days.
• Limited HSE capacity: A team of 10 HSE staff supports operations across six facilities.
• Underused video infrastructure: Existing CCTV footage is not systematically analyzed for safety events.

These delays can make it harder to preserve evidence, identify recurring hazards and communicate lessons learned quickly. They also increase the administrative workload for HSE officers, who must collect photographs, review logs, reconcile statements and prepare documentation manually.

The opportunity is to transform existing camera infrastructure into a more proactive safety-monitoring capability while retaining human oversight for critical operational and compliance decisions.

The AWS Solution: AI-Powered Safety Monitoring and Incident Reporting

Arthurite Integrated’s proposed architecture uses AWS managed services to connect video feeds, computer-vision models, incident workflows, evidence storage and AI-assisted reporting in a coordinated platform.

Camera feeds from the selected facilities are securely transmitted through site gateways to Amazon Kinesis Video Streams. Computer-vision models hosted on Amazon SageMaker analyze sampled frames to identify potential safety events, while Amazon Rekognition provides additional verification for supported detections.

When the platform confirms a relevant event, it records the detection, preserves associated evidence and routes alerts through Amazon Simple Notification Service (Amazon SNS), Amazon EventBridge and AWS Lambda. AWS Step Functions coordinates incident workflows, including acknowledgement, escalation, assignment and closure.

For documentation, Amazon Bedrock generates evidence-grounded incident summaries and draft regulatory reports. A designated HSE reviewer must review and approve every report before release or submission.

This architecture is designed to improve monitoring and response without handing safety-critical decisions entirely to automated systems.

Five Key Engineering Capabilities Behind the Proposed Platform

1. Continuous Video Analytics for PPE and Hazard Detection

The proposed platform extends the value of Ironridge’s existing CCTV infrastructure by analyzing selected camera feeds for defined safety conditions.

Custom computer-vision models on Amazon SageMaker are intended to detect:

• Missing hard hats, high-visibility vests, safety boots and gloves in designated areas.
• Unauthorized entry into restricted or permit-controlled zones.
• Visible smoke and flames.
• Liquid spills and pooling.
• Visual events that can be correlated with existing gas-detector alarms.

The solution uses activity filtering and sampled video frames rather than processing every frame continuously. This design is intended to balance detection performance with network bandwidth and cloud computing costs.

Amazon Rekognition supports verification for selected detection categories, while custom SageMaker models address the wider range of PPE items and site-specific use cases.

Models will need to be trained and validated against representative Ironridge camera imagery, including conditions such as dust, low light, glare and night operations. Accuracy targets will be agreed and measured using labeled validation data during implementation.

2. Real-Time Alerts and Automated Incident Escalation

Detecting a potential hazard is only useful when the appropriate people receive the information in time to act.

Arthurite’s proposed solution uses Amazon SNS, Amazon EventBridge, AWS Lambda and AWS Step Functions to route alerts according to event severity and coordinate the response process.

Critical events, such as visible smoke or flames and gas-detector alarms correlated with camera activity, are designed to trigger immediate notifications to relevant operational and HSE personnel. High-severity events, including restricted-zone entry and significant spills, can create incident cases and initiate escalation procedures.

The proposed targets include:

• Critical and high-severity alerts delivered within 30 seconds of confirmed detection.
• Automated escalation when alerts remain unacknowledged.
• Timestamped records of acknowledgement, assignment, action taken and closure.
• Feedback from HSE officers to help identify false positives and improve model performance.

These workflows are intended to reduce dependence on manual discovery and give safety teams a more consistent way to manage incidents from detection through resolution.

3. Generative AI for Faster Incident and Regulatory Reporting

Incident documentation often requires HSE teams to assemble information from several sources before they can produce a coherent report. This process can delay investigations and reduce the time available for preventive safety work.

The proposed platform uses Amazon Bedrock to generate structured incident summaries, preliminary cause analyses, regulatory report drafts and weekly HSE digests.

The AI system will draw on available evidence, including detection timelines, alert records, officer notes, relevant facility information and gas-detector readings. Report templates and workflow controls will help standardize documentation and identify missing information.

The proposed controls are especially important in a safety-critical environment:

• Generated reports remain drafts until an authorized person approves them.
• Preliminary cause analysis is labeled appropriately rather than presented as a confirmed conclusion.
• Reports are grounded in supplied evidence, with event references retained for traceability.
• Figures are inserted programmatically from stored data rather than invented by the language model.
• Approval records and edit histories are retained for audit purposes.

The project targets a reduction in incident report preparation from approximately six working days to less than one working day within 12 months. Regulatory report drafting is targeted at under four hours, subject to human review and approval.

These are planned performance objectives that will require validation during implementation and operation.

4. Centralized Evidence Storage and HSE Performance Dashboards

Reliable incident management requires more than alerts. HSE teams also need access to relevant evidence, historical trends and operational information.

The proposed solution uses Amazon DynamoDB to store structured detection records and incident status information, while Amazon S3 stores snapshots, evidence clips, report drafts and associated artifacts. AWS Key Management Service (AWS KMS) supports encryption, and Amazon S3 Object Lock is proposed for protecting incident evidence where retention requirements call for it.

Amazon QuickSight and Amazon Athena will support dashboards and analysis across several areas:

• Live Safety Status: Open alerts, event severity and camera health.
• PPE Compliance Trends: Compliance patterns by facility, area, shift and PPE category.
• Zone and Permit Compliance: Restricted-zone events, permit correlation and response times.
• Leak, Smoke and Spill Events: Detection records, response performance and closure status.
• Reporting Performance: Report turnaround times, review backlogs and regulatory deadlines.

Together, these capabilities are intended to help managers identify recurring hotspots, monitor response performance and use incident data to inform preventive action.

5. Secure, Resilient and Privacy-Conscious Architecture

Safety monitoring must be dependable and appropriately governed. The proposed platform incorporates security, resilience and privacy controls throughout its design.

AWS Site-to-Site VPN provides secure connectivity between Ironridge facilities and the cloud. The architecture proposes redundant VPN tunnels, local buffering during short connectivity interruptions and camera-health alerts when a stream becomes unavailable.

Amazon SageMaker inference endpoints are designed to scale with workload demand across multiple Availability Zones. The solution also uses AWS Identity and Access Management (IAM), Amazon Cognito, AWS KMS, AWS CloudTrail and Amazon CloudWatch to support access control, encryption, auditability and operational monitoring.

Privacy is another core consideration. The proposed system is designed to analyze behaviors and environmental conditions rather than identify individuals. Facial recognition and identity matching are excluded from the scope, and face blurring is an available configuration option. Video retention, evidence access and worker notification arrangements will be reviewed with Ironridge’s legal and compliance teams.

AI-assisted reporting remains subject to human approval, and the platform is not designed to shut down industrial processes, acknowledge control-system alarms automatically or replace existing gas-detection and permit-to-work systems.

Expected Business Impact: What the Project Aims to Improve

The proposed implementation establishes measurable objectives for evaluating the platform’s performance.

Performance indicator	Current baseline	Proposed 12-month target
Coverage of in-scope working hours	Less than 2%	95% or more
Alert delivery after confirmed detection	Not applicable	Under 30 seconds
PPE detection precision / recall	Not measured	At least 92% / 90%
PPE non-compliance rate	Approximately 11% in audit estimates	3%
Leak, smoke or spill detection-to-response	Approximately 4.5 hours	Under 10 minutes
Incident report preparation	Approximately 6 working days	Less than 1 working day
Regulatory report draft preparation	Approximately 3 working days	Under 4 hours, with human approval
Alerting platform availability	Not applicable	99.9%

These figures are targets documented in the proposed Statement of Work. Baselines are estimates to be validated, and actual outcomes will depend on camera suitability, network availability, model accuracy, operational adoption and other implementation factors.

If achieved, these improvements could help Ironridge identify unsafe conditions earlier, improve the consistency of incident handling and reduce the administrative burden associated with safety documentation.

A Structured 16-Week Implementation Roadmap

Arthurite has proposed a phased implementation over 16 weeks, beginning with two pilot facilities before expanding the platform across all 64 in-scope cameras.

Phase 1: Discovery and Design — Weeks 1–2

The team will assess camera placement, network bandwidth, safety policies, restricted zones, data retention and privacy requirements. This phase will establish the agreed scope and technical design.

Phase 2: Video Pipeline and Pilot Detection — Weeks 3–7

The team will configure video ingestion at two pilot facilities, prepare labeled imagery and develop initial PPE and restricted-zone detection models. Alerts and detection records will be tested before wider deployment.

Phase 3: Detection Expansion and Generative AI Reporting — Weeks 8–12

The platform will expand toward all 64 cameras, adding smoke, flame and spill detection, gas-alarm correlation, and AI-assisted incident and regulatory report drafting.

Phase 4: Dashboards, Validation and Go-Live — Weeks 13–16

The final phase will cover dashboards, model accuracy validation, load and failover testing, user acceptance testing, staff training and a shadow-mode period before production go-live.

This phased approach is intended to validate the technology in real operating conditions and give Ironridge’s HSE team the opportunity to review detections and refine thresholds before relying on alerts operationally.

Advancing AWS Solutions for Energy Operations

The Ironridge initiative illustrates how AWS cloud services, computer vision and generative AI can be combined to address operational challenges in the energy sector.

By connecting video analytics, event-driven workflows, secure evidence management and AI-assisted reporting, Arthurite Integrated’s proposed solution demonstrates an approach to modernizing HSE processes without removing human responsibility from safety-critical decisions.

The design also supports broader energy-sector use cases in which organizations need to improve operational visibility, process large volumes of information and make better use of existing infrastructure.

For Arthurite Integrated, the project represents a proposed addition to its energy technology portfolio, complementing other use cases involving smart meter analytics, AI-powered load forecasting and operational performance monitoring.

Conclusion: Turning Safety Data into More Actionable Decisions

Effective HSE management depends on identifying hazards, communicating risks and acting before incidents escalate. Manual inspections and reactive video review remain useful components of safety operations, but they can leave significant gaps when facilities operate continuously and HSE teams must cover multiple sites.

Arthurite Integrated’s proposed AWS-based platform for Ironridge Energy Operation aims to close those gaps through continuous video analytics, severity-based alerts, centralized evidence management and AI-assisted reporting.

The intended result is a more proactive safety workflow: potential hazards become visible sooner, response teams receive actionable information, and HSE professionals spend less time assembling reports manually.

Achieving these outcomes will depend on careful implementation, reliable camera feeds, representative model validation, privacy safeguards and sustained human oversight. With these foundations in place, intelligent monitoring can become a practical tool for strengthening safety operations across energy facilities.

Looking to modernize safety monitoring or operational workflows with AWS and AI?

Arthurite Integrated works on cloud architecture, AI-enabled automation and data-driven solutions designed around real operational requirements. Contact the Arthurite Integrated team to discuss how a tailored technology strategy could support your organization’s digital transformation goals.