Published October 15, 2026 | Version v1

Simulation-Based Verification of a Multimodal Sensing and Data-Integrity Pipeline for Laboratory Building-Condition Assessment: Digital-Twin Dry Run of an Arduino/ESP32 Test Programme (Code, Simulated Dataset and Report)

  • 1. University of customs and finance

Description

Description / Abstract

This record contains the software package, simulated dataset and report for a dry run of a laboratory test programme. The programme asks whether combining acoustic sensing (six SPH0645 I2S microphones, 48 kHz) with inertial sensing (eight BMI270 IMUs, 800 Hz) on Arduino Nano ESP32 nodes improves recognition of four controlled configurations (C0–C3) of a 1:20 building model. It also tests a cryptographically protected data stream: CRC-32, a SHA-256 hash chain, ECDSA P-256 signatures and Merkle-root anchoring to a ledger.

Important: no physical experiment was performed. All structural-response, recognition, synchronisation, calibration and false-alarm results come from a physically motivated modal digital twin of the test rig. They are intended for pipeline verification and planning, not as experimental evidence.

The full design of 4 configurations × 3 force levels × 2 impact points × 20 repeats (480 records in four sessions, 502 attempts including rejected impacts) was simulated. These records were then processed by the same pipeline intended for real data: an event detector, 179 acoustic and inertial features, leave-one-session-out Random Forest classification with isotonic calibration, 1000-resample cluster-bootstrap confidence intervals, and robustness tests for node loss and additive noise.

On simulated data the fusion feature set reached macro-F1 0.935 (95% CI 0.834–0.992) and the inertial set 0.911. The paired difference (ΔF1 = 0.025, 95% CI −0.003 to 0.048) did not confirm an advantage of fusion. Simulated p95 recommendation latency was 2.38 s. The simulated false-alarm rate of 5.5 events/h exceeded the 1 event/h target.

The following software checks were actually executed (on a PC, not on the ESP32):

  • 14 controlled tampering scenarios (A0–A13), all with the expected verdict. Scenario A13, a substituted physical stimulus, is shown to be undetectable by cryptography alone.
  • Verification of 288,000 stream frames.
  • A ledger-outage queue test.
  • Seven functional checks of rule-based agents for data quality, damage inference, rescue planning and human-in-the-loop coordination.

The package includes Python and Node.js source code, the run log, features, predictions, figures, and reports in English and Ukrainian. It is designed so that, after the physical sessions are completed, the simulator is replaced by a parser of real recordings and all tables and figures are recomputed.

Other Zenodo fields:

  • Resource type: Software, or Dataset with the report attached. Do not use "Journal article".
  • Keywords: structural health monitoring; multimodal sensor fusion; MEMS microphone; IMU; ESP32; digital twin; simulation; data integrity; hash chain; ECDSA; blockchain anchoring; agentic AI; human-in-the-loop; rescue planning
  • Version: 0.1-sim
  • Notes field: "Simulation dry run — no experimental measurements. To be superseded by a version containing laboratory data."

Files

README.md

Files (5.5 MB)

Name Size
md5:6bc5722d3f79f8fc908aff6787ac713e
1.3 kB Preview Download
md5:0542a1b597e20d2b00b189180653bf07
5.5 MB Preview Download