How a $2.5M R&D Grant Funded Our AI Research

European Innovation Funding for Industrial AI

sensfix has been awarded a significant research and development grant under Poland's POIR program (Programme for Intelligent Development), one of the European Union's flagship instruments for funding innovation in member states. The project carries a total value of PLN 12.8 million, with PLN 9.8 million in EU co-financing — a substantial endorsement of sensfix's technology vision and its potential to transform industrial maintenance and quality operations.

The grant period spans from January 2022 through December 2023, funding an intensive two-year research program that has accelerated our technology roadmap by years. Under the oversight of NCBiR (the National Centre for Research and Development), sensfix has used this funding to develop breakthrough capabilities in computer vision, audio AI, and multimodal decision-making that form the foundation of our commercial platform today.

Understanding the POIR Program

The POIR program is part of the broader European Structural and Investment Funds framework, designed to support research, technological development, and innovation across the European Union. Within Poland, the program is administered by NCBiR and targets companies developing technologies with high commercialization potential that address real market needs.

Winning a POIR grant is a competitive process. Applicants must demonstrate:

• Technical novelty — the proposed research must advance the state of the art, not merely apply existing technologies
• Commercial viability — there must be a credible path from research outcomes to market-ready products
• Team capability — the applicant must have the technical team and infrastructure to execute the research plan
• Market impact — the technology must address significant market demand with potential for broad adoption

sensfix's proposal was evaluated against these criteria by independent technical reviewers and was selected for funding based on the strength of our technical approach, the clarity of our commercialization strategy, and the magnitude of the market opportunity in industrial AI.

What Was Developed: Computer Vision Models

The first major research pillar funded by the grant was the development of specialized computer vision models for industrial defect detection. Unlike general-purpose image classification, industrial defect detection presents unique challenges:

• Class imbalance: Defective samples are rare compared to normal ones, making it difficult to train models with sufficient positive examples
• Subtle anomalies: Industrial defects are often minor variations — hairline cracks, slight discolorations, microscopic surface irregularities — that require high-resolution analysis and fine-grained feature extraction
• Environmental variation: Factory lighting, camera angles, surface contamination, and equipment aging all introduce variability that models must learn to handle
• Zero tolerance for false negatives: In safety-critical applications, missing a defect can have catastrophic consequences, demanding extremely high recall rates

Over the grant period, the sensfix research team developed novel architectures and training methodologies that address each of these challenges. Our models use advanced data augmentation techniques to overcome class imbalance, multi-scale feature extraction to detect anomalies across different size ranges, and domain adaptation methods to generalize across varying environmental conditions.

The result is a family of computer vision models that achieve detection rates exceeding conventional methods while maintaining the low false-positive rates required for production deployment. These models now form the core of the SAAI Suite's visual inspection capabilities.

What Was Developed: Audio AI

The second research pillar — and arguably the most innovative — was the development of audio AI models for industrial equipment diagnostics. While computer vision for manufacturing has received significant attention from the research community, acoustic diagnostics for industrial equipment remains a relatively underexplored domain.

sensfix's research focused on several key challenges:

• Noise separation: Industrial environments are loud. Extracting meaningful equipment signatures from background noise — other machines, HVAC systems, human activity, material handling — requires sophisticated signal processing and source separation techniques
• Baseline learning: Each piece of equipment has a unique acoustic fingerprint when operating normally. The system must learn what "healthy" sounds like for each individual machine before it can identify anomalies
• Fault taxonomy: Different types of mechanical faults produce different acoustic patterns. A failing bearing sounds different from a misaligned shaft, which sounds different from a loose mounting bolt. The models must not only detect anomalies but classify them by fault type to enable targeted maintenance responses
• Temporal analysis: Some faults manifest as gradual changes over time rather than sudden events. The system must track acoustic trends and detect slow degradation patterns that would be invisible in any single measurement

Our audio AI research produced models capable of detecting mechanical anomalies in compressors, motors, and rotating equipment with remarkable sensitivity — identifying faults that experienced human technicians would miss, and doing so consistently across shifts, without fatigue or distraction.

What Was Developed: The Multimodal Rule Engine

The third research pillar brought together the first two: a multimodal rule engine that combines visual and acoustic AI outputs with domain knowledge to make intelligent diagnostic and routing decisions. This is the intelligence layer that transforms raw AI detections into actionable maintenance insights.

The rule engine operates on several principles:

• Evidence fusion: When both visual and acoustic evidence is available for the same equipment, the engine combines them using Bayesian inference to produce a unified diagnosis with higher confidence than either signal alone
• Domain constraints: The engine encodes equipment-specific knowledge — for example, that a particular vibration pattern in a compressor, combined with a specific visual indicator on the housing, is diagnostic of a particular fault mode — enabling expert-level reasoning
• Severity assessment: Based on the type and magnitude of detected anomalies, the engine assigns severity scores that determine urgency and routing. Critical findings trigger immediate alerts; developing issues are scheduled for the next maintenance window
• Workflow orchestration: The engine maps diagnostic outputs to specific workflow actions — generating work orders, notifying personnel, ordering spare parts, updating maintenance records — closing the loop from detection to resolution

NCBiR Oversight and Research Rigor

As the administering body, NCBiR maintained rigorous oversight of the research program throughout the grant period. This included regular milestone reviews, technical audits, and financial accountability checks. The oversight process, while demanding, ensured that the research maintained scientific rigor and stayed aligned with the commercialization objectives outlined in the original proposal.

sensfix welcomed this oversight. It imposed a discipline on our research process that benefited the quality of the outcomes and provided external validation of our progress. Every milestone was documented, every result was measured against predefined success criteria, and every expenditure was justified against the research plan.

Accelerating the Technology Roadmap

Without the EU funding, the research outcomes described above would have taken significantly longer to achieve. The grant enabled sensfix to:

• Expand the research team by hiring specialized machine learning engineers, signal processing experts, and domain consultants during the grant period
• Invest in computing infrastructure for model training, including GPU clusters and data storage systems
• Conduct extensive field trials with industrial partners, collecting the real-world data that is essential for training production-quality AI models
• Iterate rapidly through multiple model architectures and training approaches, testing and discarding approaches that did not meet performance targets

The result is a technology platform that is years ahead of where it would have been through organic R&D investment alone. The POIR grant did not just fund research — it compressed the timeline from laboratory to production, enabling sensfix to bring AI-powered industrial solutions to market faster and with greater technical maturity.

Connection to Lodz and European Operations

The grant work was conducted primarily at sensfix's Lodz, Poland office, which serves as our European development center. Lodz has become an increasingly important technology hub in Central Europe, with a strong university system producing talented engineers and a growing ecosystem of technology companies.

For sensfix, the Lodz office is not a satellite operation — it is a core part of our technology organization. The research team in Lodz developed the fundamental AI capabilities that power every sensfix deployment worldwide. The EU grant reinforced our commitment to building deep technical capability in Europe, creating high-value research jobs, and contributing to the continent's competitiveness in industrial AI.