Practical factory AI use-cases
An Economic Times panel framed AI in factories around narrow, repeatable problems—downtime classification, tool‑life warnings, rejection-pattern summaries and mixed‑product scheduling—rather than headline GDP goals. The discussion stresses digitising decisions that managers actually make on the shop floor, not measuring flux without stabilising processes first (economictimes.indiatimes.com).
Factory artificial intelligence is showing up first in small, repeatable shop-floor decisions, not in headline promises about national output. (economictimes.indiatimes.com) In an Economic Times discussion published April 10, 2026, Vinod Kumar of PwC India and Srihari Kaninghat of JSW Group described uses such as predicting machine failures, cutting material costs and getting past pilot projects. The conversation was hosted by Anirban Chowdhury. (economictimes.indiatimes.com) The policy backdrop is bigger than one factory. India’s National Mission on Manufacturing, announced in the Union Budget 2025-26, targets a 25% manufacturing share of gross domestic product by 2035, up from 12.9% in 2023, and projects 143 million jobs. (economictimes.indiatimes.com) A year earlier, the Confederation of Indian Industry summit used a different baseline, saying manufacturing was contributing 17% of gross domestic product in March 2025. The gap in those figures reflects different measures and time periods, but both point to the same policy push for a larger factory sector. (pib.gov.in) The factory problems in this debate are concrete. Downtime classification means sorting every line stoppage into a standard reason code, so managers can see whether a loss came from a jam, a tool change, missing material or maintenance. (sgsystemsglobal.com) Predictive maintenance works like a health tracker for machines. IBM says artificial intelligence systems use live data such as vibration, temperature and pressure to forecast when intervention is needed, instead of replacing parts on a fixed calendar. (ibm.com) Tool-life warning is the same idea applied to cutting tools that slowly wear out and start producing bad parts. A December 2025 review in *Applied Sciences* said tool-wear monitoring is central to quality and productivity, but also said wider industrial use is limited by data shortages, model complexity and computing cost. (mdpi.com) Rejection-pattern summaries focus on scrap and defects. In manufacturing, a “rejection” is a part that fails quality checks, and root-cause analysis tries to connect those failures to a machine, material batch, setting or process step instead of leaving them as isolated incidents. (mdpi.com) Mixed-product scheduling is the planning problem behind one line making several versions of a product. Research on mixed-model production describes it as sequencing different models on the same line while keeping materials flowing and avoiding idle stations or bottlenecks. (strategosinc.com; utamohring.org) The panel’s argument was narrower than the usual artificial intelligence pitch: digitise decisions supervisors already make, and do it where the process repeats often enough to train a system. That is a different claim from saying sensors alone will fix a plant with unstable processes or poor data discipline. (economictimes.indiatimes.com; ibm.com) That leaves factory artificial intelligence looking less like a single moonshot and more like software for reason codes, wear alerts, defect patterns and line sequencing. The test is whether those systems move a plant from “we think we know why” to a timestamped decision someone can actually act on. (economictimes.indiatimes.com; sgsystemsglobal.com)
