CondiCloud is the commercial outcome of doctoral research into a fundamental problem in condition monitoring: how to make intelligent fault detection work without expert knowledge, fault data, or enterprise infrastructure.
Vibration-based condition monitoring is well-established in research and in large industrial settings. The unsolved problem was making it work for everyone else.
For decades, condition monitoring research has produced increasingly sophisticated methods for detecting faults in rotating machinery — bearing defects, gear wear, rotor imbalance, and more. The academic literature on vibration analysis, signal processing, and machine learning applied to fault detection is extensive and mature.
Yet in practice, these advances have remained confined to large industrial organisations. The gap between what research has made possible and what most manufacturers can actually access is wide — and widening. Three specific barriers have kept that gap open:
University of Auckland
Faculty of Engineering
The core research contribution is a condition monitoring framework that addresses all three barriers simultaneously — without trading one off against the others.
Rather than requiring documented failure records, the system learns what healthy operation looks like from a brief period of normal running. It then detects statistically significant changes from that learned baseline — entirely without any fault examples in training. Any operating machine can provide the data it needs, immediately.
Vibration signals carry a large amount of information, not all of it equally useful. The system automatically identifies which aspects of the signal are most diagnostic for that specific machine — without any manual configuration, expert input, or parameter tuning. It adapts to each asset independently.
The same vibration signal is analysed from several complementary perspectives simultaneously. This makes the detection robust across different fault types and operating conditions — where a single-perspective approach would miss faults that only become visible from a different analytical angle.
Each component addresses one of the three deployment barriers. Learning from healthy data eliminates the fault data requirement. Self-configuration eliminates the expert requirement. The edge-cloud architecture eliminates the infrastructure cost barrier. Together, they produce a system that can be deployed on day one, on any rotating machine, by a maintenance technician — without any of the preconditions that have historically blocked adoption.
The research contribution is not any single technique in isolation — the individual analytical components draw on well-established foundations in the condition monitoring literature. The contribution is their integration into a unified, self-contained pipeline that satisfies the practical constraints that existing approaches do not: no failure history, no specialist, no proprietary infrastructure.
The research programme includes both benchmark dataset validation and laboratory experiments designed to reflect real deployment conditions.
The detection algorithm was evaluated against multiple widely-used public benchmark datasets in the condition monitoring community — covering bearings and gearboxes under a range of operating conditions, fault types, and severity levels.
These datasets are the standard reference points used by the research community to compare detection methods. Evaluation against them establishes that the algorithm's performance is comparable to — and in several cases exceeds — supervised approaches that require labelled fault data.
Across 32 independent test assets, the system demonstrates consistent early fault detection without any fault examples in training — validating the core research hypothesis.
Beyond benchmark performance, the research includes controlled laboratory experiments that validate the complete system — not just the algorithm — under conditions that mirror real deployment.
Experiments use interchangeable healthy and failed gearboxes to establish controlled ground truth, tested across a matrix of operating speeds and loads. Two independent monitoring nodes run simultaneously on the same equipment to validate multi-node consistency — a critical requirement for reliable deployment.
False alarm testing under changing operating conditions — varying load and speed on a healthy machine — validates that the system does not generate spurious alerts when conditions change, which is the most common cause of alert fatigue in deployed monitoring systems.
The research addresses a gap that the condition monitoring literature has identified but not resolved — making the technology practically accessible, not just technically capable.
A validated condition monitoring framework that demonstrably eliminates the three primary deployment barriers simultaneously — without sacrificing detection performance relative to approaches that require labelled fault data.
A complete system — algorithm, edge computing architecture, and cloud platform — that can be deployed by a maintenance technician on day one, on any rotating machine, at a cost accessible to small and medium manufacturers.
Democratising access to technology that has meaningful economic impact — reducing unplanned downtime, extending equipment life, and improving maintenance efficiency — for the organisations that most need it and have historically been excluded from it.
Whether you're an academic, an engineer evaluating the system, or a manufacturer curious about what's under the hood — we're happy to discuss the research in detail.