Project CHAKSHU RPMP

Condition-Based & Predictive Maintenance

Catch the failure before it grounds the fleet.

The Rotorcraft Predictive Maintenance Platform brings health monitoring and remaining-life prediction to helicopters separated by half a century of avionics — from first-generation analogue airframes to current-generation, HUMS-equipped types — behind one type-agnostic model.

See the approach → Delivery roadmap
◎ OSA-CBM / ISO 13374 · 6-layer pipeline ● Live
RAW SIGNAL → FEATURE → STATE → HEALTH → RUL → ADVISORY L1 DATA ACQUIRE L2 FEATURE CIs L3 STATE DETECT L4 HEALTH FUSE L5 RUL PROGNOSE L6 ADVISE ACTION INPUTS → L1 HUMS vibration oil debris / SOAP usage manual / retrofit COMPONENT Main gearbox · RT-15 HEALTH 0.82 · CAUTION RUL ESTIMATE 128 ± 22 hrs ADVISORY: INSPECT @ 100 hrs
“The expert in maintenance does not wait for the part to fail — he reads the warning the machine has already given.”
RPMP design principle · condition over calendar
3
Eras served
6
CBM layers
0%
Traceable
0
Cloud / airgapped

The Challenge · One fleet, three eras

The legacy fleet can’t tell you it’s sick.

Indian rotary-wing operators fly aircraft separated by fifty years of avionics. First-generation types are analogue airframes with fixed time-between-overhaul limits and no onboard health monitoring — large, ageing fleets carrying high maintenance burden. Current-generation types are modern airframes whose variants carry a Health & Usage Monitoring System. One platform must serve a data-rich modern type and data-poor legacy types without forcing them into the same assumptions.

0
Onboard sensing (legacy)

First-generation analogue types have instruments only — no recorded health data.

TBO
Fixed-interval limits

Maintenance driven by hours and calendar, not by observed condition.

~350
Ageing airframes

A large legacy fleet beyond its design life, being progressively replaced.

HUMS
Modern HUMS data

Multi-channel vibration, usage and exceedance data ready to exploit.

The Approach · OSA-CBM / ISO 13374

Six layers, one certifiable chain.

RPMP follows the internationally recognised six-layer condition-monitoring reference model. Each layer is a single-responsibility service: data flows forward as immutable events, every stage enriches the picture, and condition-monitoring layers stay dormant for tails that lack the input data — without affecting the rest.

L1Data Acquisition

Acquire

Receive and normalise every source — HUMS files, engine/usage data, manual captures, retrofit feeds — into one validated record.

L2Data Manipulation

Extract features

Signal processing into condition indicators — gear/bearing CIs, oil-debris trends, rotor-balance coefficients, usage severity.

L3State Detection

Detect

Compare features to limits and per-tail baselines; emit nominal / caution / warning states and anomalies.

L4Health Assessment

Assess health

Fuse states into a single, explainable component health index that always resolves back to its evidence.

L5RUL

Prognose

Estimate remaining useful life — physics + data fused — always with a quantified uncertainty band.

L6Advisory Generation

Advise

Turn health and RUL into a recommended action and horizon — inspect, limit or replace. Advisory only; humans decide.

Capabilities

What the platform watches.

Every modality is mapped to the failure modes it detects, and every output is traceable to the evidence and algorithm version that produced it — the property that makes analytics-driven decisions acceptable for airworthiness review.

Vibration & condition indicators

Order and spectral analysis yields gear and bearing indicators — FM0, FM4, NA4, sideband energy and bearing defect frequencies — from raw vibration.

Gearbox · rotor · shafts · bearings

Oil debris & SOAP

Chip-detector and lab oil analysis quantify ferrous and non-ferrous wear — the strongest early warning when fused with vibration.

Available to all types · incl. legacy

Rotor track & balance

Derives blade adjustments to minimise vibration — computed automatically from HUMS, or from portable-analyser surveys on legacy types.

1/rev & n/rev · track split

Usage & regime recognition

Classifies flight into severity regimes so fatigue life is credited by what the aircraft actually did — not just hours flown.

Safe-life consumption · severity

RUL with uncertainty

Physics-based fatigue and data-driven degradation fused into a remaining-life estimate — published with a confidence band, never a bare number.

Hybrid digital twin · UQ

Explainable & traceable

Every assessment lists its drivers and links to source data, so a certifying engineer can audit it. Human-in-the-loop throughout.

Airworthiness-ready evidence

Inside the algorithms

From a raw waveform to a defensible decision.

Diagnosis extracts condition indicators from vibration, oil and usage data; prognosis projects them to a remaining life with quantified uncertainty. Every method below is industry-standard for rotorcraft transmissions, and every output is traceable to the evidence that produced it.

Diagnostics — detect

Layers 2–3

Turn raw signals into condition indicators, then into a state.

Time-synchronous averagingFM0 / FM4 / NA4M6A / M8AOrder trackingSideband analysisEnvelope · BPFO/BPFISpectral kurtosisOil debris / SOAP

Detection · fixed limits · per-tail SPC · Mahalanobis · isolation forest · autoencoder · fleet-relative outlier

Prognostics — predict

Layer 5 · RUL

Fuse physics and data into a remaining-life distribution.

Paris-law crack growthKalman / EKF / UKFParticle filterGaussian processLSTM / Bi-LSTMSimilarity-basedBayesian hybrid

Every estimate is a distribution — the advisory acts on a conservative quantile, with SHAP-style driver attributions.

Physics / model-based

Needs no failure history

Crack-growth and cumulative-damage models give a life estimate from first principles — ideal for the data-poor legacy fleet.

State-space filtering

Tracks the degradation state

Particle and Kalman filters carry a full posterior, handling non-linear, non-Gaussian wear.

Data-driven

Learns from the fleet

Gaussian processes and recurrent networks sharpen the estimate as run-to-failure history accumulates.

Legacy Hybrid Strategy

Make the data-poor fleet a first-class citizen.

Because the legacy fleet has no onboard sensing, RPMP never assumes a vibration feed exists. Data availability is a per-tail property, and the platform degrades gracefully: usage-based prognostics work with nothing more than flight logs and oil-sample results, while full condition monitoring engages wherever HUMS or a retrofit unit is present.

A

Mobile, offline-first field capture

Ground crew record usage, a guided regime questionnaire, oil-sample results and NDT findings — structured data that would otherwise live only on paper.

B

Optional retrofit data-acquisition unit

A small, non-intrusive DAU with a few accelerometers and a chip-detector tap presents the same internal record as HUMS — no downstream change.

C

Usage-based prognostics first

With little failure history, physics and usage models lead; data-driven models earn trust as evidence accumulates, and uncertainty tightens over time.

Per-tail data availability graceful degrade
MODERN (HUMS) full condition monitoring LEGACY (ANALOGUE) manual capture + RETROFIT DAU optional sensing UNIFIED RECORD source-agnostic

Data management

Sovereign data, governed for the life of the fleet.

Predictive maintenance lives or dies on data. RPMP treats every signal, sample and record as a controlled asset with an owner, a lifecycle and a retention rule — so the analytics are reproducible and the evidence holds up at airworthiness review.

Master & reference

One governed asset model

Tail, component, configuration and limits as master data with a single owner — type-agnostic, effectivity-resolved, never duplicated.

Time-series lifecycle

Hot → warm → cold → archive

High-rate vibration and feature streams are tiered and down-sampled by age, balancing query speed against decades of retention.

Quality & lineage

Reproducible & provable

Schema-on-write validation, quarantine for bad records, and immutable provenance from every feature back to its source file and algorithm version.

Records & retention

Kept as airworthiness records

Safety-relevant data is retained for the life of the aircraft plus the statutory margin, under a tamper-evident, policy-driven schedule.

Security & classification

Classified & airgapped

Every data domain is classified (including SCOMET-controlled data) with role- and attribute-based access; no data leaves the enclave.

Interoperability

Standards, not silos

Exchanges align to the S-Series ILS specifications and the MIMOSA asset model, so RPMP plugs into enterprise logistics rather than walling it off.

S1000D · tech pubsS2000M · materialS3000L · LSAS5000F · in-service feedbackMIMOSA OSA-EAIDAMA data qualitySCOMET

The Application · Built for the hangar

Easy to use — even under time pressure.

The interface favours clarity over density: a small number of well-signposted screens, plain-language status, evidence one click away, and no action that cannot be undone or audited. Three capabilities anchor it.

Fleet availability board

Availability is computed from open items and life-limit margins — never hand-set — so the board can’t drift from reality.

AC-1043Serviceable
AC-2261Limited · MEL
AC-0789AOG

Incident follow-ups

An incident is a case with one owner, a status lifecycle and an append-only follow-up thread — nothing in the history can be silently edited.

Open→Investigating→Resolved
CASE-4471 · gearbox caution

Follow-up · oil sample requested · assigned to MRO planner · SLA 48h

Inventory & spares

Rotables and consumables are tracked against forecast demand from advisories, so a shortfall is flagged before it grounds a tail.

Tail rotor bearingstock 4
MGB chip detectorlow · 1
Engine hot-section moduleon order

Incident workflow · tracked like a board

From a health alert to a closed case.

Every flagged anomaly becomes a card that moves across a board — reported, analysed, assigned, discussed, followed up and resolved — with an append-only message thread that survives audit. Several cases sit on the board; the lifecycle below plays out on one of them.

◎ Live board · 8 open cases ● Live board
HEALTH REPORT ANALYSIS ASSIGN THREAD FOLLOW-UP RESOLUTION CASE-4490 Tail-rotor 1/rev ↑ CASE-4456 Engine N2 droop CASE-4501 MGB chip light CASE-4438 MGB oil debris ↑ CASE-4421 Vibration trend CASE-4402 LLP margin low CASE-4399 RTB completed CASE-4471 MGB · caution NA4 ↑ · RUL 128h M ▸ CASE-4471 · CONVERSATION 8 updates · append-only ◎ RPMP · Health bot · T+0 NA4 on main gearbox RT-15 crossed caution — RUL 128 ± 22 h. RK Reliability · T+0:42 Sidebands and oil debris agree — bearing wear, not a gear fault. MP MRO planner · T+1:10 Raising work order WO-2207. Need an MGB chip detector — any stock? SS Supply · T+1:18 1 serviceable at base, 2 inbound — reserved one against WO-2207. RK Reliability · T+1:30 Inspect at 100 h; oil sample requested to confirm the wear mode. ◎ RPMP · Scheduler · T+1:35 Slot reserved Tue 09:00 · tail flagged Limited (MEL). QA Quality · T+2:05 Approved — close on a chip-clear and a normal oil result. ✓ RPMP · T+3:40 · RESOLVED Chip detector cleared, oil normal — CASE-4471 closed, audit sealed.
● health / resolution ● in-progress step append-only · fully audited

Beyond the aircraft · Fleet & logistics

Manage the whole fleet, not just the airframe.

The same asset model and the same predictive signals run the sustainment loop — every aircraft, every part, every order and every sortie. A predicted failure becomes a planned part, a reserved hangar slot and a re-tasked aircraft, instead of a surprise grounding.

01

Fleet asset management

Every tail, base, configuration and life-margin in one registry — with basing, transfers and modification state tracked.

RCM · FMECA · Weibull

02

Spares & inventory

A rotable pool and multi-echelon stock sized to a fleet-availability target — not blanket stock levels.

METRIC / VARI-METRIC · (s,S) · lateral resupply

03

Procurement & orders

Auto requisitions from reorder points and forecast shortfalls; a full, audited PO lifecycle with AOG escalation.

requisition → PO → received · repair orders

04

Routing & scheduling

Tail assignment and maintenance-slot scheduling as one optimisation; predicted life enters as a soft due-date.

tail assignment · time-space network · multi-criteria

◎ The sustainment loop — one decision, not three ● Closed loop
HEALTH / RUL predict SCHEDULE reserve slot SPARES / ORDER fill the gap AVAILABILITY fly it feedback · fly → sense → predict

Delivery Approach · Sequenced in steps

Skeleton, then flesh-out, then production.

Each capability is taken through three maturity stages — a walking skeleton that runs end-to-end on stubs, then real algorithms and screens, then hardening for performance, safety and security. Nine phases run against explicit exit gates; the skeleton gate is the one that matters early.

P0Start

Setup & governance

Charter, environments, standards and the definition of done.

P1Discover

Domain modelling

Fleet, maintenance cycle, physics and regulatory model.

P2Skeleton

Architecture & skeleton

A walking skeleton through all six layers on synthetic data.

P3Data

Data foundation

Modern HUMS + legacy hybrid capture, validated and lineage-tracked.

P4Monitor

Condition monitoring

Feature extraction, state detection and health assessment live.

P5Predict

Prognostics & RUL

Physics + data-driven remaining-life with uncertainty.

P6Apply

App, follow-ups, fleet

Dashboard, incident follow-up workspace and availability board.

P7–8Production

V&V & deploy

Evidence package, then airgapped rollout, training and sustainment.

Cross-cutting throughout: security & data governance · MLOps · quality & airworthiness · change & training

Built to recognised condition-monitoring & airworthiness standards

ISO 13374OSA-CBMMIMOSA OSA-EAI MSG-3 (rotorcraft)Safe-life / damage-tolerance CEMILACDO-178C / DO-254SCOMET

Atmanirbhar Bharat · Sovereign sustainment

Keep the fleet flying on your own data.

A full technical dossier — the activity & phase breakdown and the technical design — is available under briefing. Request access to discuss a deployment.

Request the dossier → Back to capabilities