DEV Community

Nashtarin Nur
Nashtarin Nur

Posted on Edited on

The Brownfield Edge Overlay: Architecting a 4-Engine Pipeline for Legacy Operational Technology

The majority of value in industrial environments exist in brownfield environments with old hardware. From heavy manufacturing, food processing, and bulk logistics, many core processes use old PLCs, bespoke serial data buses and isolated electro-mechanical systems.

Full hardware upgrades are costly and can cause disruption. Adding generic cloud IoT gateways can introduce significant security risks and issues with protocol conversion. To overcome this, non-Invasive Physical AI can be used at the edge of the system to observe, reason and act.

Four Engine Architecture

To modernize legacy operational technology, a solution should be implemented that can observe the environment in high fidelity, process the data locally and integrate with existing control systems. This can be achieved using a 4 Engine Architecture for Physical AI in brownfield environments.

  1. Identification Engine – External Visibility

Legacy machinery rarely emits structured data. A non-invasive identification layer should be implemented which includes UWB anchors, RFID gates, BLE beacons and vision systems. This creates an accurate sub-meter spatial map of the environment identifying workpieces, mobile carriers and legacy machinery.

  1. Sensing Engine – Non-Invasive Sensing

Instead of connecting directly to the PLCs to gather diagnostic data, sensing should be done externally. Sensors can be attached to the machinery to capture high frequency data. This includes audio, tri-axial vibration, temperature and current draw. The data is normalized at the edge.

  1. AI Decision Engine – Baseline Modelling for Legacy Machinery

Each piece of legacy machinery will have different wear characteristics and calibration drift. This means generic models in the cloud will not understand them. Localised models are implemented based on the unique characteristics of the machinery being upgraded. The normalized data from the edge is processed to give a risk score based on historical degradation curves and spatial context. This identifies issues such as bearing wear, tool misalignment and bottlenecks days in advance.

  1. Action Engine – Controlled Actions

The insights from the machinery will only provide value if it can be acted upon. The action engine connects the edge intelligence with the legacy infrastructure. All actions are run in a controlled manner. This includes sending standard analogue/discrete signals to the PLC, adjusting variable frequency drives and creating prioritized maintenance tickets.

Controlled Operation

When implementing autonomous systems to work with machinery that may be decades old, it is important to maintain control. Many legacy systems do not have modern diagnostic capabilities so sending an uncontrolled command could cause damage to the machinery or risk operator safety if the edge models misidentify the unusual state.

A Physical AI system within an enterprise will always ensure the system is working within a controlled envelope. This is done using 3 levels of risk:

Autonomous – Low Risk

The system will act autonomously to complete non-invasive actions. This includes logging degradation data, updating spatial work in progress and queuing up replacement spares.

Semi-Autonomous – Medium Risk

Actions are taken based on context but will stop to await engineer approval. This includes changing conveyor speeds, curing times and planned maintenance.

Controlled – High Risk

Hard coded interlocks are used to ensure the system does not enter any algorithmic loops. This will stop the machinery safely using standard emergency relays.

To leverage the untapped potential in brownfield industries, deterministic intelligence can be added at the edge. If you are an engineering leader or technical founder looking to explore architectural concepts for brownfield modernization and deep tech ventures, you can find out more about our research here at Aperture Venture Studio.

Top comments (0)