IMU-Based Monitoring for HVAC Systems

Project Context

This engagement represents a typical early-stage exploration project focused on monitoring and diagnostics for HVAC systems.

The client’s objective was to better understand the mechanical behavior of an existing HVAC installation, with the longer-term goal of detecting abnormal conditions and anticipating failures. While the idea was conceptually sound, it was unclear whether low-cost embedded sensors could reliably capture meaningful data in real-world operating conditions.

At this stage, the project was exploratory by nature. The main risk was investing too early in an architecture or sensing strategy that would not yield actionable insights.

Initial Situation

The initial concept relied on the assumption that vibration data could be used as a proxy for system health. However, several questions remained unanswered:

• Which physical phenomena were actually observable
• Whether vibration signatures were stable over time
• How sensitive the measurements would be to mounting, orientation, and location
• Whether environmental noise would dominate the signal

The challenge was not building a monitoring device, but determining whether the data itself would be meaningful.

Technical Exploration Phase

Atallis proposed an exploration phase centered around a 9-DOF IMU, allowing simultaneous measurement of acceleration, rotation, and orientation.

Development kits were used to:

• experiment with different mounting locations
• evaluate sensor orientation sensitivity
• capture raw data across multiple operating modes

The goal was not immediate analysis or classification, but data characterization: identifying repeatable patterns, noise sources, and limits of observability.

Iterative Learning

Early tests quickly revealed that several assumptions were overly optimistic. Some vibration components initially thought to be significant were highly dependent on installation details, while others were masked by external factors such as airflow, structural resonance, or transient operating states.

Through iterative testing:

• sensor placement was refined
• sampling strategies were adjusted
• irrelevant dimensions were deprioritized

This phase helped narrow the problem space and prevented the project from chasing misleading signals.

Architecture Clarification

As the sensing strategy became clearer, the system architecture could be refined with confidence:

• only the necessary sensing dimensions were retained
• data rates were aligned with real signal bandwidth
• firmware complexity was kept proportional to actual needs

At this point, technical decisions were driven by observed behavior, not theoretical expectations.

Outcome

By the end of the exploration phase:

• the feasibility of IMU-based monitoring was clearly established
• limitations and edge cases were explicitly identified
• the project transitioned from hypothesis-driven to evidence-driven

Most importantly, the client could decide how to proceed based on real data, rather than assumptions.

Ongoing Role of Atallis

Following the initial exploration, Atallis continued to support the project by:

• refining embedded firmware for robustness
• guiding data interpretation strategies
• advising on future extensions, including anomaly detection

Atallis’ role remained focused on reducing uncertainty and ensuring that each technical step was justified by measurable results.

What This Case Illustrates

This case illustrates the importance of validating sensing strategies early, especially in systems where signal quality, noise, and installation variability can significantly impact outcomes.

By prioritizing learning over early optimization, teams can avoid building sophisticated systems on unreliable data foundations.

Note: This case study is intentionally anonymized and based on real projects completed by Atallis. Certain details have been simplified or combined to respect confidentiality commitments.