From R&D to Patent Acquisition: Why Companies Are Changing Their Innovation Strategy

Commercial and industrial buildings consume enormous amounts of electricity, but the tools available to monitor and manage that consumption are fundamentally inadequate. A traditional electricity meter tells a building owner only the total units consumed — nothing about which machine, equipment, or system is responsible for the bill. Existing solutions to this problem — such as smart meters or sub-metering — either provide only marginally better aggregate data (smart meters) or require an expensive, invasive sensor installation on every single electrical circuit or appliance in the building (sub-metering). For a typical industrial facility with 400–500 pieces of equipment across multiple distribution panels, sub-metering requires hundreds of individual sensors, extensive wiring work, significant capital expenditure, and ongoing maintenance — making meaningful energy management economically unviable for most commercial and industrial users.

The invention is a device and method that achieves individual appliance-level energy monitoring from a single measurement point — the building's main electrical panel — without any sensors on individual appliances. The device captures the combined electrical waveform of the entire building at 4 million samples per second. Because every electrical appliance has a unique power consumption signature — a distinctive pattern of voltage, current, harmonics, and switching behaviour when it turns on or off — the device's embedded machine learning engine can identify each appliance and attribute its consumption to it in real time. The ML model is self-learning: it trains on historical data from the specific building it is deployed in, continuously improving its identification accuracy over time and recognising new appliances as they are introduced. The system also predicts future consumption, detects equipment degradation before it causes failure, and generates prioritised recommendations for energy saving — all from a single installation requiring approximately five minutes with no disruption to building operations.

Yes. Prior art in Non-Intrusive Load Monitoring (NILM) existed primarily in the residential sector, where a small number of low-power appliances with repetitive and well-defined signatures made single-point disaggregation feasible. No prior art had successfully commercialised NILM for commercial and industrial buildings — environments characterised by hundreds or thousands of diverse, high-power assets operating simultaneously, with complex overlapping electrical signatures including industrial motors, compressors, HVAC chillers, and variable-frequency drives. The prior art gap was the absence of an ML model capable of reliably disaggregating aggregate waveforms in this level of electrical complexity. Existing commercial solutions either addressed only residential loads, required per-circuit physical sensors (abandoning the non-intrusive principle entirely), or provided only aggregate demand data without asset-level attribution. This invention is the first commercially deployed and patent-protected solution that closes this gap: a self-learning ML engine validated at India's largest industrial and pharmaceutical enterprises — including Maruti Suzuki India Limited and Biocon Limited — achieving 98.24% appliance identification accuracy in complex commercial and industrial environments from a single measurement point.