Improving safety using deep learning

Maersk Drilling is constantly striving to increase safety and efficiency. That’s why they wanted to implement a conditional red zone monitoring solution on the Maersk Discoverer rig. Working with FMJ Rolloos, we developed and implemented a vision-based AI solution to alert crews in case of unsafe situations.

A drill floor is a hazardous area

With multiple people working in a small (hazardous) space surrounded by heavy equipment, drilling operations can be tough. Drilling crews and operators are used to working with so-called red zones. These zones are high-risk areas on the drill floor but despite this, the drilling crew still needs to enter these zones in order to perform specific tasks. This obviously leads to contradictory operational instructions.

Challenge

The challenge in this project was twofold: technical and operational. Technology can only be a tool when improving operational safety. It depends whether people are engaging in it if the desired improvements can actually be realized. For this project the key elements are accurately tracking people and its integration with clear operational procedures.

Accurately tracking people

To alert people in case of an unsafe situation it is essential to track people and equipment very accurately. Operation is outdoors, offshore without stable connectivity to a cloud, within steel structures, 24/7, day and night. There is a great variety of people, i.e. shapes, color coveralls, permanent/temporary crew, and equipment blocking lines of sight, i.e. (temporary) equipment moving in and out.

Clear operational procedures

In order to define safe and unsafe behavior it is key to have clear operational procedures. As the drilling process involves a dynamic operation the unsafe areas move around and change based on the state of the operation. This basically creates conditional red zones; which dynamics, i.e. size, shape, conditionality, need to be defined in procedures.

Solution: embedded safety & efficiency

When developing the conditional red zone monitoring solution we have worked closely together with the crew, engineers and management of the Maersk Discoverer and FMJ Rolloos. Key to the success of the project was to create a modular tool to enforce continuous and iterative operational perfomance.

Process
optimization

Optimize operational procedures based on actual human performance.

Red zone
monitoring

Alerting the drill crew when a red zone is breached; accurately and real time.

Performance
enhancement

Standardized statistics on human behavior and operational conditions.

Technoloy

The solution we developed combines operational management, robust edge infrastructure and artificial intelligence. It consists of the following five elements.

Visualization of procedures

We upgraded the common textual working procedures by a more storybook-like version. We took the analogy of a coaching board in team sports to visualize the sequence of operational steps.

We created a workshop for crew and management to map all actions carried out during operation. This way we identified and mapped all the hazards. A side benefit was that this also proved to be an easy way to standardize operations. It increased quality of work by making procedures clear for everybody. And the procedures could now be used for training purposes and as baselines for the red zone monitoring module.

High-performance edge infrastructure

We developed and implemented a dedicated edge infrastructure. As it involved an ATEX/IECEx certified system to be installed offshore, special requirements applied concerning hardware.

    • Special low-latency, high-resolution, moulded CCTV cameras were fitted in an explosion protected housing
    • A series of Nvidia GPUs processed all live video in real time
    • The Polaris edge node integrated and processed all data from video, sensors and operational data
    • The Polaris platform enabled edge-to-cloud monitoring and data access
    • In compliance with offshore and explosion protection specification

Deep learning people tracking

Our deep-learning algorithms perform very accurate and fast people detection. With a combined performance of over 99% and a latency of close to 40-50ms we outperform most advanced operational solutions.

The observation area is monitored by multiple CCTV cameras. The feeds and analyses from these cameras are then stitched into a single result. This requires to be done in real time and has to be robust to prevent any mutliplication of detections.

Real-time alerting user interface

The operator of the drilling process, i.e. the driller, uses our Red zone monitoring interface to monitor the live operation. With a top-down view of the drill floor he is able to spot all the people, including those in the blind spots.

One or multiple conditional red zones are either active (red/on) or deactivated (green/off). If a person enters an active red zone, an alarm is raised. This is both audible and visual, to the driller and the crew out side on the drill floor.

We designed this interface to also include functionalities for system health monitoring and configuration.

Post-event analytics management dashboard

Key to the project was to improve operational performance and close the management feedback loop. Via the Polaris edge-to-cloud platform the system is able to securely synchronize all local data offshore to a central cloud database onshore. We integrated data from our CCTV cameras, existing equipment sensors and operational data.

An insights dashboarding tool allows operational management to explore, analyse and benchmark their crews and rigs. This web browser based tool integrates the insights from multiple rigs into a single dashboard. It combines common KPIs with with analyses on human behavior and performance.

Result

The most important result of the project was the increased situational awareness and safety of the crew on the drill floor. The simple user interface, powerful analytics and actionable insights support both crew and management.

Visual indication of red light makes it clear for personnel on the rig floor as to when the red zone is not accessible. This removes any uncertainty they may have.
Driller, Maersk Discoverer

Fast & accurate
analytics

Our artificial intelligence performed fast and accurately. The inference took 40-50ms. The entire time between a person being captured by the CCTV cameras and the alarm being raised took close to 150ms. The modular system setup allowed for quick optimization processes for new installations at other rigs.

Robust edge
infrastructure

The Polaris edge infrastructure operated well 24/7 within a salty offshore environment. It proved its agility to integrate a variety of existing industrial systems and their protocols. Polaris' edge-to-cloud synchronization was optimized for low-bandwidth satellite communication.

Actionable
insights

The integration of multiple data sources plus the red zone analytics made it possible to generate integrated operational insights. The real-time situational awareness increased via the conditional red zone lights plus the alerting mechanism. The collected data allowed for taking post-event optimization actions.

Solution partner

Want to learn more?

If you want to learn more about this case or what we can do for you, contact us today.