Chilling Centre Telemetry: Choosing Between 4G, LoRaWAN and a 5G Pitch
Kolhapur Sahakari Dudh Sangh collects about 42,000 litres a day through 60 village societies. Three bulk milk coolers must hold milk at 4 °C; a rise of 2 °C sustained for 40 minutes sours the whole batch. Today a worker reads each dial every two hours and writes it in a register. The plant has 2 Mbps ADSL. The two remote chilling centres, 11 km away, have no wired link and patchy coverage - roughly two bars of 4G, no 5G. Mains power fails six to ten times a month. Vendor A quotes internet-connected temperature loggers with 4G SIMs at ₹4,500 per unit plus ₹200 per unit per year in data charges. Vendor B quotes LoRaWAN loggers with a gateway rated for 6 km: higher one-time cost, no recurring SIM charge. Management wants the board to approve "5G with AI" and also wants an assurance that milk-pricing and member data stay private.
1. Choose the connectivity for the loggers. Justify the choice using range, payload size, power and recurring cost, and state the one design feature the power situation forces.
Read the worked answer
Payload first. A temperature reading every five or ten minutes is tens of bytes. Bandwidth is irrelevant here; 5G's gigabit rates and 1 ms latency buy nothing. The 2 Mbps ADSL is likewise not the bottleneck. The decision therefore rests on range, penetration, battery life and recurring cost.
4G (Vendor A). Ready-made network, no gateway to site, but reception of two bars implies marginal signal, and a metal-clad chilling room will worsen it. Recurring cost is small - nine units × ₹200 ≈ ₹1,800 a year - so cost is not the differentiator; coverage reliability is. A dropped link during a 40-minute excursion is exactly the failure that matters.
LoRaWAN (Vendor B). The 6 km figure is line-of-sight; 11 km to a centre means the gateway must sit on high ground between sites or a second gateway is needed, which must be costed, not assumed. In exchange, sub-GHz LoRa penetrates walls better, node radios sleep for years on one battery, and no SIM means no recurring charge and no dependency on a telecom's billing.
Recommendation. Hybrid: LoRaWAN for sites with clear line of sight to a gateway, 4G only for the site that lacks it.
Forced feature. With six to ten outages a month, each logger must buffer at least 24 hours locally and back-fill on reconnect, or the record shows flat lines during precisely the outages when a cooler warms up.
2. Management asks for "5G with AI". Assess that request and propose a realistic sequence of steps with a justification for each.
Read the worked answer
5G. 5G in India is deployed in bands that need small cells every few hundred metres in dense areas; rural Kolhapur has none, and the centre already reports no 5G at all. Approving "5G" would pay for capability the problem does not consume: a sensor sending a 20-byte reading needs neither high throughput nor low latency. The honest answer to the board is that 5G is not a prerequisite and its absence blocks nothing.
AI. Spoilage prediction needs a labelled history - timestamped temperature traces matched to accepted and rejected batches. The register holds two-hourly manual readings, which cannot label a 40-minute excursion; two-hourly sampling would have missed the event entirely. Training a model on that data produces confident nonsense.
Realistic sequence.
- Instrument the coolers at 5–10 minute intervals and store the raw series (this makes every later step possible).
- Add a deterministic rule alert first: "temperature above 6 °C for 20 minutes" or "rising 1 °C in 30 minutes" → SMS to the supervisor. Rules catch the great majority of excursions, cost little, and are auditable.
- After three to four months, when labelled data exists, evaluate anomaly detection against the rule baseline and adopt it only if it beats the rule on missed events.
This order spends money on the sensing layer that generates value today, and defers the model until there is data to justify it.
3. The co-op is nervous about the cloud. Recommend a service model for the telemetry and dashboard, say what must stay on-premises, and name the legal hook that makes the data handling question more than commercial preference.
Read the worked answer
Service model. Telemetry ingestion suits PaaS: a managed IoT broker, time-series store and rules/alerts engine, with no server for the dairy to patch - decisive, since the co-op has no system administrator. The phone and browser dashboard suits SaaS, so alerts reach supervisors without custom front-end work. IaaS would be chosen only if the co-op wanted full control, which it does not have the staff to exercise. A private or region-locked public cloud with encryption at rest and in transit satisfies the privacy worry while keeping managed services.
What stays on-premises. Member records, milk-pricing data and payment information are commercially sensitive and unrelated to the control loop; keep them in the plant's own database and push only temperature telemetry and equipment IDs to the cloud. If member personal data ever does leave, it becomes "sensitive personal data" under Section 43A of the IT Act, 2000, which requires a body corporate handling such data to maintain reasonable security practices and exposes it to compensation for negligence - so a written agreement with the provider, a defined data location and a retention period belong in the contract, not just in the vendor's slide deck.
Design note. Alerts must fire from the cloud rules engine to the supervisor's phone, but a local rule on the gateway should also trip a siren, so an internet outage never leaves a cooler unmonitored.