From Industrial AI Pilots to Real Business Value: How ABB Is Connecting Predictive AI and IoT

From Industrial AI Pilots to Real Business Value: How ABB Is Connecting Predictive AI and IoT

Industrial artificial intelligence is moving beyond experimental applications and becoming part of everyday manufacturing and process operations. However, achieving measurable returns requires more than installing connected sensors or deploying predictive algorithms. It depends on integrating operational data, engineering knowledge, automation systems and human decision-making into a coordinated digital infrastructure.

Rajesh Ramachandran, Global Chief Digital Officer for ABB's Automation business area, believes that combining Industrial Internet of Things (IIoT) technologies with predictive AI can help manufacturers improve asset reliability, optimise energy consumption and move towards increasingly autonomous operations.

His perspective highlights an important challenge facing industrial companies: turning large volumes of operational data into decisions that produce measurable improvements.

1. Industrial AI and IoT: Moving Beyond Data Collection

The Industrial Internet of Things has become a major driver of digital transformation. Connected sensors, intelligent instruments and networked control systems allow manufacturers to collect operational information continuously. Meanwhile, predictive AI provides the analytical capabilities needed to identify patterns, detect abnormal conditions and anticipate potential equipment failures.

These technologies serve different but complementary purposes. IoT provides visibility into physical equipment and processes, while AI interprets the collected information and identifies potential operational risks.

However, collecting more data does not automatically improve industrial performance. A vibration reading or temperature measurement has limited practical value without understanding the equipment's operating conditions, maintenance history and engineering limits.

In my view, the next stage of industrial digitalisation should focus less on increasing the volume of collected data and more on improving its engineering relevance. A smaller amount of well-contextualised information can support better maintenance decisions than millions of disconnected sensor readings.

2. Data Quality and Engineering Context Determine AI Performance

According to Ramachandran, a successful predictive AI strategy should begin with a clearly defined business problem.

Industrial companies need to identify what they want to predict, how early a warning is required and which operational decision will benefit from that prediction.

This approach places engineering requirements ahead of technology selection.

For predictive models to generate useful results, information from multiple sources must be connected, including:

  • Operational Technology (OT): Real-time measurements from sensors, PLCs, DCS platforms and other control systems.
  • Information Technology (IT): Enterprise information, production records and business-related operational data.
  • Engineering Technology (ET): Equipment specifications, process knowledge, operating limits and engineering documentation.
  • Maintenance Records: Equipment service histories, inspection results, failure reports and previous corrective actions.

The integration of OT, IT and ET provides a broader understanding of equipment behaviour.

For example, a sudden increase in motor vibration may indicate a developing mechanical problem. However, the interpretation changes depending on motor speed, loading conditions, bearing history and recent maintenance activities.

This is why industrial AI requires more than data science. Engineering knowledge provides the context that allows analytical results to become meaningful operational information.

3. Integrating Predictive AI into Existing Industrial Systems

One of the biggest obstacles to industrial AI adoption is not necessarily model development. It is integrating AI-generated insights into existing operational workflows.

A predictive model may identify an abnormal condition with high accuracy, but its value remains limited if the maintenance team does not receive the information in time or cannot determine what action to take.

Ramachandran emphasises that AI solutions must integrate securely with existing systems and involve clear responsibilities for acting on their recommendations.

For industrial facilities, this requires attention to several practical considerations:

  • Integration with existing automation and monitoring infrastructure.
  • Secure communication between operational and enterprise systems.
  • Appropriate access to real-time and historical equipment data.
  • Clear alarm prioritisation and maintenance notification procedures.
  • Defined responsibilities for reviewing and approving AI recommendations.
  • Continuous evaluation of model performance and operational results.

From an automation engineering perspective, AI should complement existing control and protection architectures rather than introduce unnecessary complexity into them.

Predictive analytics can support maintenance planning and operational optimisation, but safety-related control functions and critical protection mechanisms must retain their established engineering boundaries.

4. ABB Genix: Building an Industrial AIoT Platform

ABB is addressing these integration challenges through its ABB Genix Industrial IoT and AI platform.

The platform is designed to connect industrial data, contextualise information and apply analytics across different operational layers. Its purpose is to help organisations improve efficiency, asset reliability and sustainability through more coordinated use of industrial information.

A platform-based approach is particularly relevant for large industrial facilities where equipment, control systems and maintenance applications often operate in separate information environments.

Instead of deploying isolated AI applications for individual machines, manufacturers can work towards a more integrated architecture that supports data ingestion, integration, analysis and optimisation.

This also creates opportunities to extend successful applications across multiple production lines and facilities.

However, platform deployment alone does not guarantee operational improvement. Data governance, system integration, engineering validation and workforce adoption remain necessary parts of the implementation process.

5. Escaping Pilot Purgatory: Scaling AI Across Industrial Operations

Many manufacturers have successfully demonstrated AI applications on individual machines or production processes. The more difficult task is extending those results across an entire plant or enterprise.

This situation is often described as pilot purgatory: promising digital projects remain isolated demonstrations instead of becoming established operational capabilities.

Ramachandran identifies organisational leadership as an important factor in overcoming this problem. Companies need clear business objectives, suitable digital infrastructure and cooperation between technology teams and operational personnel.

In practical terms, scaling industrial AI requires organisations to address several challenges:

  • Data Silos: Information must be accessible across relevant operational and business systems.
  • Legacy Infrastructure: Existing PLC, DCS, SCADA and asset management environments may require carefully planned integration.
  • Workforce Adoption: Engineers and maintenance personnel need to understand how AI recommendations should influence their work.
  • Operational Ownership: Every AI application needs clearly defined responsibility for reviewing its outputs and initiating actions.
  • Deployment Strategy: Successful applications should be expanded according to demonstrated operational value rather than technology availability alone.

I believe manufacturers should avoid treating AI adoption as a race to automate every process. A more practical strategy is to identify recurring operational problems, validate solutions in controlled environments and expand applications only when their benefits can be demonstrated.

6. Measuring Industrial AI ROI Through Operational Performance

For industrial applications, AI performance should not be evaluated through model accuracy alone.

A technically impressive algorithm does not necessarily create business value. Its contribution must be reflected in measurable improvements such as equipment availability, production yield, maintenance costs, energy consumption and operational safety.

A useful starting point is to establish an operational baseline before introducing predictive AI.

For example, if a predictive maintenance system provides sufficient warning to prevent an unexpected equipment shutdown, the financial benefit can be evaluated against avoided production losses, emergency maintenance expenses and potential secondary equipment damage.

ABB has highlighted an application involving multivariate anomaly detection within its Genix Asset Performance Management solution.

According to Ramachandran, recent customer deployments have identified potential equipment failures six to ten days before they occur. This warning period can give maintenance teams additional time to investigate abnormal conditions and arrange corrective interventions.

Nevertheless, prediction quality must be evaluated carefully.

False positives can lead to unnecessary inspections and maintenance activities, while missed warnings can expose production equipment to significant operational risks.

For this reason, industrial AI performance should be assessed through both technical indicators and actual maintenance outcomes.

7. Predictive Analytics, Prescriptive Intelligence and Digital Twins

Industrial AI is gradually moving beyond identifying what might happen towards explaining why an event is likely to occur and recommending what should be done next.

Predictive analytics identifies potential future conditions. Prescriptive analytics evaluates possible responses, while causal analysis can help investigate the underlying factors contributing to abnormal behaviour.

Digital twins extend these capabilities by creating virtual representations of physical assets, processes or systems. When connected with real-time operational data, simulation models can help engineers evaluate different operating scenarios before making changes to physical equipment.

This combination offers several potential benefits:

  • Earlier identification of abnormal operating conditions.
  • Improved understanding of potential equipment failure mechanisms.
  • Evaluation of alternative maintenance or operating strategies.
  • Reduced uncertainty when assessing proposed process changes.
  • Better-informed decisions based on operational scenarios.

In my opinion, the combination of predictive analytics and digital twins could become particularly valuable for complex industrial processes. Rather than simply notifying engineers that a parameter has exceeded its normal range, future systems should increasingly help explain the possible causes and evaluate the consequences of different corrective actions.

8. Agentic AI and the Future of Autonomous Industrial Operations

Looking ahead, Ramachandran expects predictive, prescriptive and generative AI capabilities to work increasingly closely together, with agentic AI helping connect analytical insights to operational workflows.

Predictive AI can identify emerging risks. Generative AI can help engineers interpret technical information, while agentic AI can support workflow execution and coordinated tasks within defined operational boundaries.

This development could change how industrial personnel interact with automation systems.

Instead of manually reviewing multiple dashboards, maintenance records and engineering documents, personnel may increasingly use AI-assisted systems to investigate equipment conditions, identify relevant information and prepare recommended actions.

However, greater automation does not mean removing human responsibility.

In industrial environments, decisions involving safety, equipment integrity and production continuity must remain subject to appropriate engineering oversight. AI-generated recommendations require validation, and automated actions must operate within clearly established permissions and system constraints.

The objective should not simply be to create factories that use more AI. It should be to develop industrial operations that make better decisions, respond more effectively to changing conditions and maintain appropriate levels of human accountability.

9. Engineering Perspective: Industrial AI Success Starts with Operational Discipline

ABB's approach highlights a broader lesson for manufacturers: industrial AI is not an isolated software investment. It is an extension of the existing automation, information and engineering infrastructure.

The strongest opportunities are likely to emerge where companies already have identifiable operational problems, sufficient historical and real-time data, and teams capable of translating analytical findings into practical actions.

For automation engineers, this means that the development of AI-enabled operations will increasingly involve cooperation between control engineering, instrumentation, IT infrastructure, maintenance and data science.

It also means that existing industrial assets should not be overlooked. Many established facilities already contain valuable operational information within PLCs, DCS platforms, condition monitoring systems and maintenance databases. Connecting and contextualising these resources may provide a more practical starting point than replacing existing infrastructure.

Ultimately, the transition towards intelligent industrial operations will depend on how effectively companies combine digital technologies with engineering experience. AI can improve visibility and decision-making, but measurable industrial value still comes from solving real operational problems.

From Industrial AI Pilots to Real Business Value: How ABB Is Connecting Predictive AI and IoT
Previous post