What are the most significant risks that could jeopardize the success of the project? How are these risks being managed? What contingency plans are ready?
One of the most significant risks at this stage of the project is ensuring that the dataset we collect from the HVAC ducts is large and diverse enough to train a reliable machine learning model. Because we decided to build our own dataset rather than rely solely on existing research datasets, the success of our model now depends heavily on the quality and quantity of the acoustic data we collect. If the dataset is too small or lacks variation, the SVM classifier may overfit and fail to generalize to new ducts. To mitigate this risk, we are structuring our dataset carefully by collecting baseline recordings from healthy ducts and then introducing controlled mechanical defects in stages (minor, moderate, and severe). This staged approach allows us to capture progressive structural degradation and increases the diversity of the training data. As our worst case contingency plan, we would need to supplement with pre-existing structural health monitoring datasets. However, this is the worst-case scenario because the data in these datasets might vary significantly from what we are aiming to test on and therefore might make our model more complicated.
Another risk is the integration of the hardware components, particularly ensuring that the sensors, tapping mechanism, and Jetson-based computing platform can reliably capture and process acoustic signals. If the hardware components are not synchronized properly or if the signals contain excessive noise, this could affect both the data collection and the real-time analysis. To manage this risk, we are testing the signal processing pipeline early using simulated or preliminary data and researching amplifier and sensor configurations that will improve signal quality. Additionally, we are reviewing testing methodologies from relevant research papers to guide our experimental setup. As a contingency plan, we can simplify the real-time processing requirements by performing some of the feature extraction offline if we experience performance limitations.
Were any changes made to the existing design of the system (requirements, block diagram, system spec, etc)? Why was this change necessary, what costs does the change incur, and how will these costs be mitigated going forward?
We made some refinements with our system design clarifying how we will structure the defect categories used for training. Rather than using random or unstructured damage, the dataset will now be organized into controlled defect severity levels such as healthy, minor damage, moderate damage, and severe damage. This change improves the interpretability of the model and ensures that the classifier learns meaningful acoustic signatures associated with progressive damage. The cost incurred from these changes is the additional time and effort required to physically create defects in the ducts and collect labeled data. We plan to mitigate these costs by standardizing the testing procedure and collecting multiple recordings per defect stage so that we can generate a sufficiently large dataset from a limited number of duct samples in the coming week.
Provide an updated schedule if changes have occurred.
The overall schedule of the project remains mostly consistent, but the timeline for dataset development and testing has been adjusted slightly. Now that the HVAC ducts and parts have been obtained, the next phase of the project will focus on collecting training data from intact ducts and validating the signal processing pipeline using this data. During the same period, the team will continue integrating the embedded system components that will eventually automate the data collection process.
Part A: Global Factors
LeakLink addresses a global need for more efficient and scalable inspection of HVAC duct systems, which are widely used in commercial buildings, hospitals, schools, and industrial facilities worldwide. In many regions, HVAC infrastructure is aging, and regular inspection is required to maintain indoor air quality and system efficiency. However, current inspection methods often rely on manual labor and expensive specialized equipment. Our solution provides a lower-cost and more automated alternative that can help facility managers detect structural damage such as cracks or corrosion earlier. This is especially relevant in large urban environments where buildings contain extensive duct networks that are difficult to inspect manually. By enabling faster and more frequent inspections, the system can support preventative maintenance and reduce long-term infrastructure costs globally.
Part B: Cultural Factors
In many building maintenance settings, technicians rely heavily on practical experience and quick diagnostic methods rather than complex analytical tools. Our system is designed with this in mind by providing clear classification outputs that can be easily interpreted by technicians without requiring extensive training in machine learning or signal processing. Also, this system will not replace the need for HVAC technicians and engineers, but rather feed into the narrative of human-machine collaboration. Additionally, maintenance practices vary between regions and organizations as some emphasize preventative inspections while others focus on reactive repairs. By providing an automated tool that identifies damage early, our system can help encourage preventative maintenance practices and sustainable building maintenance practices by reducing energy consumption and minimizing the material waste associated with replacing damaged HVAC components. This aligns with cultural trends around energy efficiency and climate responsibility, especially in cities pushing for greener buildings.
Part C: Environmental Factors
HVAC systems are one of the largest contributors to energy consumption in commercial and institutional buildings. Structural damage such as cracks and corrosion in ductwork can lead to significant energy loss due to inefficient airflow and pressure drops. LeakLink aims to detect these issues early so that repairs can be made before energy inefficiencies worsen. By enabling earlier detection of duct degradation, LeakLink can reduce wasted energy and improve overall HVAC efficiency. Additionally, identifying damage before it becomes severe can extend the lifespan of duct infrastructure, reducing the need for large-scale replacements and minimizing material waste. Currently, metal HVAC ducts are not routinely inspected due to the time and effort it takes to disassemble panels to perform checks, leading to the entire system worsening over time and needing to be replaced. Furthermore, LeakLink will provide the exact location of the defect whereas normal tests just indicate that a defect exists. LeakLink will perform regular system checks so only individual panels that are identified as degraded will need to be replaced. This supports more sustainable building maintenance practices and reduces the environmental impact associated with HVAC system failures.
A was written by Mahati, B was written by Rayann, and C was written by Adithi
