Engineering labs in India often focus on confirming known results rather than teaching students how to fix broken machines. Students frequently adjust data to meet grading thresholds instead of learning to diagnose real hardware failures. This diagnostic gap leaves graduates unprepared for the complex challenges they face in professional engineering.

Learning to diagnose and set right failures may be the most important experiment missing from Engineering labs

Across UG Engineering laboratories in India, practical coursework is built to confirm known results, not to investigate unknown ones. Instruments are calibrated before students arrive. Test rigs use components chosen because they behave. The lab manual is a checklist that terminates at an answer already printed in the syllabus. Students power on the supply, take six predictable readings, plot a curve that reproduces the theoretical derivation, and leave. When a reading drifts or refuses to settle, the unwritten convention -- tolerated by instructors running back-to-back batches of 60 or more -- is to nudge a potentiometer, discard the outlier, or quietly adjust the numbers until they clear the grading threshold.

The result is a graduate fluent in how idealised systems behave and inexperienced with actual hardware. Someone has to isolate a physical defect the textbook never modelled; usually under time pressure, with incomplete information. This diagnostic deficit is not an accident of pedagogy. It is an economic consequence of how mass technical education is assessed. Deterministic labs scale cheaply: checking whether a student's calculated output lands within 5% of the expected value takes a grader seconds. Judging how that same student traces an intermittent cold solder joint or a clogged sensor port requires an experienced instructor watching an investigation unfold in real time; a resource a department running a 100 students per practical slot does not have.

The incentives that follow are perverse. A student who reaches the target value by swapping working components until the circuit accidentally functions receives full marks. A student who finds a blown protection diode, spends the session tracing the fault to its physical root cause, and documents the breakdown carefully is penalised for an incomplete dataset. Taught this way, Engineering students learn that an anomaly is something to conceal.

Computational tools have widened the gap rather than closed it. An undergraduate can now generate firmware, synthesise digital logic, and run a multiphysics thermal model in an afternoon. Physical debugging has not accelerated to match because no simulation removes contact resistance, component ageing, parasitic inductance, or thermal expansion from a real assembly. In product development and manufacturing, much of an engineer's economic value sits precisely in that gap: resolving the divergence between a clean simulation and hardware that refuses to cooperate.

Fixing this does not require large capital outlays on imported equipment. It requires using existing infrastructure -- and instructor time -- differently. Three models are practical now.

Shared fault-injection testbeds: Instead of buying dozens of identical benches that sit idle most of the week, a department can put a small number of instrumented rigs -- a fluid loop, a thermal exchanger, a motor test stand -- on the campus network. Instructors inject faults programmatically: a supply-rail sag, a partial blockage upstream of a pressure sensor. Students query the live sensor streams remotely and are graded on the diagnostic log they produce -- the hypotheses formed, tests run, and the root cause isolated -- not on whether the rig yielded a textbook curve. As the fault injection is software and scheduling rather than hardware, the marginal cost per student falls as utilisation rises.

Field deployment in unstructured environments: Practical modules should push telemetry nodes into the campus's own working infrastructure: the utility pump house, a rooftop solar array, a wastewater line. Failure modes that never appear on a clean workbench show up within days -- radio attenuation behind a concrete wall, sensor drift on a sagging battery, thermal throttling inside an unventilated enclosure. This costs almost nothing in equipment. What it demands is negotiation with the estates department and a willingness to grade messy data.

None of this is at odds with policy. AICTE's Examination Reform Policy (2018) explicitly directs institutions to assess at the higher levels of Bloom's taxonomy. Diagnostic labs are a direct implementation of that mandate in the one part of the curriculum where it is currently ignored. The industrial timing matters too. Industries cannot be staffed by graduates who freeze when the bench misbehaves. An unexpected measurement is not an error to hide. In most of Engineering, it is where the actual work begins.

The writer is Chief Operating Officer and National Coordinator at I-STEM, under the Office of the Principal Scientific Adviser, Government of India.