Recovering ₹2.6–3.2 Lakhs/Year from a 250 kWp Manufacturing & Warehouse Rooftop Plant — Central India

Background

This case study covers an anonymized manufacturing & distribution facility in Central India operating a 250 kWp rooftop solar plant commissioned under a CAPEX model roughly three years before the audit. The plant was sized to offset the facility's daytime load — shift operations, warehouse HVAC, material-handling equipment — with surplus exported to the local DISCOM under the prevailing net-metering arrangement.

Central India sits in a moderate-irradiance zone (annual GHI ~4.8–5.2 kWh/m²/day, materially below Rajasthan's 5.5–5.8 but above the Indo-Gangetic plain). At the operator's prevailing industrial C&I tariff (midpoint ~₹8.00/kWh per plant_profile row), the plant was projected to deliver an annual generation of approximately 3,75,000 kWh and offset roughly ₹30 lakhs in annual electricity costs, with a payback window of under two years. The capital cost and OEM warranty terms were average for the commissioning period.

After two operating cycles, the facility's plant manager had two unresolved questions: (1) generation was visibly below the original projection in some months, but the variability looked plausible ("maybe monsoon? maybe dust?"), and (2) the facility's daytime load had grown since commissioning, but there was no way to tell whether the rooftop plant was actually meeting the new demand or whether the surplus/deficit shape had shifted in a way the original feasibility had not modelled. The OEM dashboard remained the only monitoring layer, and as at the hospital precedent, it provided generation figures without a benchmarked "what should have happened" overlay.

The Challenge

The operator's situation before onboarding to BijleeAI was structurally identical to the hospital case study in its surface symptoms (generation below projection, no quantified loss figure) but materially different in two ways that shaped the analysis: (a) the daytime load was industrial, with a strong midday peak and intermittent demand from material-handling equipment, and (b) the regulator (DISCOM) applied a net-metering settlement that did not pay retail-rate for exported units — so every kWh the plant failed to self-consume was structurally worth less than the inverter display figure suggested.

• No cleaning schedule tied to PR drift: The plant's cleaning cadence was event-driven but reactive — the operator called the cleaning contractor when the panels looked dirty, usually once per quarter. There was no connection between the cleaning decision and the cumulative revenue loss from soiling on that particular rooftop. On a 250 kWp plant, the gap between a 4-week cleaning interval and an 8-week interval is roughly ₹15,000–20,000/month in lost generation revenue (estimate).
• No self-consumption visibility: The facility's energy management stopped at the inverter boundary. Whatever load was running behind the meter was invisible to the solar monitoring layer, so the operator could not see whether the rooftop plant was genuinely offsetting the midday peak or silently exporting at a non-retail rate.
• No downtime quantification at the string level: The inverter display aggregated all string-level output into a single number. There was no way to detect a single-string outage (whether from soiling on a specific roof slope, a wiring fault, or a combiner-box issue) until cumulative monthly generation was visibly below expectation — and by that point a month of revenue had already been lost.

What BijleeAI Found

When BijleeAI ingested the plant's 14-month history of monthly_readings and joined it to the satellite-derived irradiance data for the Central India location (Open-Meteo historical archive — no API key, refreshed monthly via the jobs/weather-refresh.js cron), the monitoring engine identified three distinct loss channels rather than the single soiling narrative the hospital case study surfaced.

• Soiling losses quantified at ₹1.2–1.8 lakhs/year (estimate): By normalising actual generation against weather-adjusted expectation, the engine isolated soiling as a ~5–7% annual loss, mapping cleanly onto the uncleaned weeks in the early-March-to-mid-May window and again in October–November. At the operator's industrial tariff, the midpoint ₹1.5 lakhs/year translates to ~₹12,500/month — the figure anchored in the Revenue-at-Risk KPI above.
• String-level downtime contributed ~₹0.6–0.9 lakhs/year (estimate): By comparing per-month generation variance against the expected shape derived from irradiance, the engine flagged two string-level anomalies that the inverter-level summary had hidden — one combiner-box issue in month 7 (~₹18,000 lost in that month alone, ~₹0.25 lakh/yr prorated) and a partial shading issue from a new HVAC unit installed on the adjacent rooftop in month 11 (~₹0.4 lakh/yr).
• Self-consumption gap of ~25–30% (estimate): BijleeAI's load-matching analysis — built from the facility's hourly consumption data uploaded as part of the audit — showed that roughly a quarter of the plant's midday generation was exporting to the DISCOM at the net-metered rate rather than offsetting retail-rate load. Two adjustments (a 90 kW midday shift in the warehouse HVAC schedule, and a 40 kW timing change on the material-handling cycles) recovered ~₹40,000–60,000/year with zero capex.

The 60 kWp expansion headroom identified (vs the hospital's 30 kWp) reflects the larger unused rooftop area at this site — the facility has two adjacent warehouse blocks, only one of which is currently solarised. The expansion financial model built from actual post-audit PR (not the original feasibility's assumed PR) yields the 1.7-year payback figure in the KPI block.

The Solution

The BijleeAI deployment combined the same five platform levers the hospital case study uses, but reframed for a manufacturing-rooftop context — where downtime detection and self-consumption recovery are higher-value than the hospital's expansion-led orientation.

• Cleaning ROI alerts (event-driven, not calendar-driven): The platform monitors the rolling-30-day PR drift and triggers a cleaning alert only when the estimated revenue loss from continued soiling exceeds the cleaning cost threshold (~₹8,000–12,000 per event for a 250 kWp plant in this region). This replaced the quarterly visit-on-demand cadence with economically-optimised scheduling.
• Loss-waterfall diagnostics per month: Each month's loss waterfall (soiling / thermal / clipping / downtime / wiring) is generated from monthly_readings + Open-Meteo irradiance — no on-site sensors required. For a manufacturing client this surface area matters more than for a hospital because downtime categories dominate loss attribution here, not soiling.
• String-level anomaly detection: A monthly anomaly scan flags per-string performance deviations against the irradiance-expected shape. This is what surfaced the combiner-box and partial-shading issues above — neither would have shown on the OEM dashboard until the cumulative monthly number was visibly off-trend.
• Self-consumption uplift advisory: The hourly-load vs hourly-generation overlay generates quarterly recommendations for shifting flexible loads into the midday solar window. For this facility the two actionable shifts recovered ~₹50,000/year (estimate, midpoint of ₹40–60k) without any new equipment.
• Expansion feasibility model: Built bottom-up from the plant's actual post-audit PR (not the EPC feasibility assumption) and the operator's prevailing tariff, the 60 kWp expansion model shows a 1.7-year payback — making it a straightforward investment-committee conversation.

Results — Before / After

The before/after picture (drawn from event_history_log cleaning events joined to monthly_readings by month_year):

The Cleaning ROI on this plant comes out lower than the hospital's 3–5x because the absolute soiling loss is smaller (Central India is dustier than the Indo-Gangetic plain but milder than Rajasthan), while the cleaning cost per event is the same per kWp. The case still pays out — at 2.5–4x with a cleaning spend of ₹25,000–30,000/yr, the recovered revenue is ₹62,500–₹1,20,000/yr (estimate) — but the headline ratio sits a notch below the hospital case. This is a useful piece of context to share with prospects: the ROI band moves with climate, not with the platform.

Recommendation (concrete next step)

This quarter, the operator should approve three BijleeAI-mediated actions, in this order:

• Lock the cleaning cadence to the BijleeAI alert feed — target 4–6 cleaning events/year, event-triggered not quarterly. At the operator's prevailing industrial tariff this alone recovers an estimated ₹0.9–1.2 lakhs/yr of the ₹2.6–3.2 lakhs/yr recovery path.
• Implement the two midday load-shifts identified in the self-consumption analysis (warehouse HVAC schedule + material-handling timing) — zero capex, ~₹0.4–0.6 lakhs/yr uplift (estimate), 30-day payback.
• Take the 60 kWp expansion to the investment committee using the BijleeAI-built model — 1.7-year payback at the current tariff, materially stronger than the original EPC feasibility assumption.

Track these three via the monthly anomaly-scan report (already running on the platform). If the underlying KPIs — PR drift, self-consumption ratio, and downtime-flag count — do not move in the indicated direction within 90 days, the platform surfaces that as a flag in the next quarter's review, not as a silent miss.