RopeVision: Cost Effective Solution to Detect Lift Rope Defects Using AI (REF:C-0061)
E&M InnoPortal Trial Project Ref. No.:

Overview

The RopeVision system operates as an integrated, edge‑based diagnostic platform, purpose‑built for autonomous use inside lift machine rooms. Unlike conventional sensors that capture data intermittently, it combines five distinct hardware modules to align high‑speed visual telemetry with accurate spatial mapping.

Core Hardware Modules：
1. Localized Processing Unit (PU):Housed in a wall‑mounted industrial enclosure, the PU acts as the computational hub, running Python‑driven AI routines. On‑site analysis removes the energy cost and latency of cloud processing while ensuring data sovereignty.
2. High‑Speed Video Capture: An industrial camera records at 750 fps with 640 × 480 resolution. This frame rate preserves sub‑millimeter clarity even when the cabin reaches 10 m/s.
3. Event‑Triggered Illumination: A powerful LED delivers synchronized bursts of light during image capture. The targeted illumination prevents motion blur and stabilizes RGB levels across the rope surface, regardless of shaft lighting conditions.
4. 1D LiDAR Telemetry: Positioned at the rope entry, the LiDAR projects a beam onto the car roof to measure displacement with 1 mm precision. Sampling at 2 Hz, it provides reference vertical coordinates for each frame.
5. Real‑Time Status Indicator: A tri‑color signal unit provides immediate feedback, showing green for normal operation, yellow for alerts, and red for required action, all determined by the daily Health Index.

Problem Addressed Traditional rope health assessment is labour-intensive and time-consuming, and manual counting of broken wires is prone to human error. An effective detection method is therefore necessary to maintain a consistent safety baseline.
Innovation CNN-Based Structural Defect Detection and Health Index Computation
Defect identification employs a two-tier classification architecture that combines deterministic colour-space analysis with a lightweight CNN. This hybrid approach leverages the strengths of each method: colour-range mapping provides computationally efficient and highly interpretable rust quantification, while the CNN extends detection capability to structural defects that cannot be identified through colour alone.

Tier 1: Colour-Range Rust Detection
The foundational rust detection algorithm analyses each masked pixel by mapping its RGB values to a predefined colour range corresponding to the spectral characteristics of iron oxide deposits. The rust colour profile was established with reference to documented rust colour coordinates and validated against field samples collected during the initial proof-of-concept trial. When a pixel's colour values fall within the defined rust range, it is flagged as a rust-positive pixel and marked with a green overlay dot for visual verification.

Tier 2: CNN-Based Structural Defect Detection
While colour-range analysis is effective for surface corrosion, structural anomalies such as broken wires and excessive grease deposits require pattern recognition beyond chromatic features. To address this, a purpose-built lightweight CNN was developed and deployed on the edge processing unit to classify masked rope-surface patches into defect categories.
Key Benefits • Low cost: At an installation cost of approximately 2% of total lift value
• Server connection is not required: RopeVision has moved away from power-intensive cloud processing models. With its localized edge computing unit, the system can operate without internet access, thereby minimizing cybersecurity risks.
• Easy to install: The major works can be carried out during planned maintenance. No additional lift suspension time will be required.
Patent and Award - Short-term Patent: HK30127485
- Inventions Geneva 2025 (Gold Medal)
Project Reference Civil Aid Service Headquarters