Data from: Application of piezoelectric intelligent materials in pipa adaptive tuning system and its influence on performance stability
The primary article associated with this dataset has been retracted.
Data files
Dec 11, 2025 version files 47.06 KB
-
NFIS_training_data_2000_rows.csv
45.68 KB
-
README.md
1.38 KB
Abstract
A dataset of 2,000 samples was collected, including voltage (0.1–5 V), temperature (10–40°C), and humidity (20–90% RH) values, along with corresponding output adjustments. The NFIS utilised Gaussian membership functions to categorise sensor inputs into linguistic terms (e.g., “High Voltage,” “Medium Temperature”), and a comprehensive rule base of 40 rules was established for adaptive tuning. Training of the NFIS was conducted using gradient-descent backpropagation with a learning rate of 0.01 and L2 regularisation, validated through 5-fold cross-validation. Real-time performance data was transmitted via an ESP32 microcontroller to an AWS IoT Core database, with user adjustments and data visualisation provided through a mobile application.
https://doi.org/10.5061/dryad.12jm63z87
Description of the data and file structure
A dataset of 2,000 samples was collected, including voltage (0.1–5 V), temperature (10–40 °C), and humidity (20–90 % RH) values, along with corresponding output adjustments. The NFIS utilised Gaussian membership functions to categorise sensor inputs into linguistic terms (e.g., “High Voltage,” “Medium Temperature”), and a comprehensive rule base of 40 rules was established for adaptive tuning. Training of the NFIS was conducted using gradient-descent backpropagation with a learning rate of 0.01 and L2 regularisation, validated through 5-fold cross-validation. Real-time performance data was transmitted via an ESP32 microcontroller to an AWS IoT Core database, with user adjustments and data visualisation provided through a mobile application.
Code/software
Python, all code available on github: https://github.com/Zikou80/Pipa_Tuning_Stability
Access information
Other publicly accessible locations of the data:
Materials used in the study included; Lead Zirconate Titanate (PZT) Sensors (Model PZT-5H, APC International, USA); Light Aluminum Non-Invasive Adjustable Attachment Brackets (Misumi Corporation, Japan); Velcro Straps (Velcro USA Inc., USA); Shielded Flexible Cables (Alpha Wire, USA); Cable Ties or Clips (Panduit Corporation, USA); DS18B20 Digital Temperature Sensors (Maxim Integrated, USA); Torque Wrench (Tohnichi, Japan); 24-bit ADC Module (Model ADS1256, Texas Instruments, USA); Butterworth Low-Pass Filter (Custom component, configured with parts from Texas Instruments, USA); High-Pass Filter (Custom component, configured with parts from Texas Instruments, USA); 0.1 µF Ceramic Capacitors (Murata Manufacturing Co., Ltd., Japan); 10 µF Electrolytic Capacitors (Nichicon Corporation, Japan)
A dataset of 2,000 samples was collected, including voltage (0.1–5 V), temperature (10–40°C), and humidity (20–90% RH) values, along with corresponding output adjustments. The NFIS utilised Gaussian membership functions to categorise sensor inputs into linguistic terms (e.g., “High Voltage,” “Medium Temperature”), and a comprehensive rule base of 40 rules was established for adaptive tuning. Training of the NFIS was conducted using gradient-descent backpropagation with a learning rate of 0.01 and L2 regularisation, validated through 5-fold cross-validation. Real-time performance data was transmitted via an ESP32 microcontroller to an AWS IoT Core database, with user adjustments and data visualisation provided through a mobile application.