https://doi.org/10.5061/dryad.hhmgqnkrh

Description of the data and file structure

The algorithms used for the comparative analysis of an RNN versus IIR digital filters are contained in two compressed files in .rar format. Each file includes the datasets used, which consist of pH measurements for both ideal and real scenarios, corresponding to the absence and presence of mechanical disturbances, respectively. The files also contain the algorithms used for their training, analysis and application.

Files and variables

RAR File 1: RNN.rar

Description:

This .rar file contains the results of the Recurrent Neural Network (RNN) for filtering perturbations in pH measurements. Inside the file, you’ll find the datasets used, the Python source code for training the RNN, notebooks with result analysis, and graphs showing the network’s performance.

.csv files with training data: The CSV files provided in the folder correspond to data collected from various pH experiments under different conditions. The naming convention follows the structure:

Perturbation Data (with disturbances):

• "1_pH4_BL_Golpes_ON.csv"
• "2_pH4_BL_Agitador_ON.csv"
• "3_pH4_Mult_Golpes_ON.csv"
• "4_pH4_Mult_Agitador_ON.csv"
• "5_pH7_BL_Golpes_ON.csv"
• "6_pH7_BL_Agitador_ON.csv"
• "7_pH7_Mult_Golpes_ON.csv"
• "8_pH7_Mult_Agitador_ON.csv"
• "9_pH10_BL_Golpes_ON.csv"
• "10_pH10_BL_Agitador_ON.csv"
• "11_pH10_Mult_Golpes_ON.csv"
• "12_pH10_Mult_Agitador_ON.csv"

• "13_pHCoke_BL_Agitador_ON.csv"

• "14_pH7-pH4_BL_Agitador_ON.csv"

• "15_pH7-pH4_BL_Golpes_ON.csv"

No Perturbation Data (without disturbances):

• "1_pH4_BL_Golpes_OFF.csv"
• "2_pH4_BL_Agitador_OFF.csv"
• "3_pH4_Mult_Golpes_OFF.csv"
• "4_pH4_Mult_Agitador_OFF.csv"
• "5_pH7_BL_Golpes_OFF.csv"
• "6_pH7_BL_Agitador_OFF.csv"
• "7_pH7_Mult_Golpes_OFF.csv"
• "8_pH7_Mult_Agitador_OFF.csv"
• "9_pH10_BL_Golpes_OFF.csv"
• "10_pH10_BL_Agitador_OFF.csv"
• "11_pH10_Mult_Golpes_OFF.csv"
• "12_pH10_Mult_Agitador_OFF.csv"
• "13_pHCoke_BL_Agitador_OFF.csv"
• "14_pH7-pH4_BL_Agitador_OFF.csv"
• "15_pH7-pH4_BL_Golpes_OFF.csv"

Each file name is composed of five parts:

Number: The first digit in the file name represents the index of the experiment (1, 2, 3, etc.). It is a sequential identifier.

pH Level: The next part, such as pH4, pH7, or pH10, indicates the pH level used in the experiment. “pH4-pH7” refers to alternating measurements between these two pH levels.

Measuring instrument:

• BL: Refers to measurements taken by a microcontroller.
• Mult: Refers to measurements taken by a multimeter.

Perturbation Type:

• Golpes (Spanish for “Hits”): Physical impacts were applied to the system.
• Agitador (Spanish for “Agitator”): Mechanical agitation was applied to the system.

State:

• ON: Indicates the perturbation was active during the experiment.
• OFF: Indicates the perturbation was not active during the experiment.

Each CSV file contains three columns with data from the experiments, structured as follows:

1. Temporal sample: Instead of exact timestamps, temporal samples are recorded.
2. Voltage: The voltage measured in the system, which is correlated with the behavior of the pH sensor.
3. pH: The pH level measured at each point in time.

“testtFINAL.csv”: During the training phase, different neural network configurations were explored. This file contains the results of each tested model. Below is a description of the columns:

• run: Identifier of the run.
• operation: Type of operation.
• started: Start date.
• Time: Duration of the training.
• label: Information about the trained model.
• l1: Number of neurons in the first layer.
• l2: Number of neurons in the second layer.
• l3: Number of neurons in the third layer.
• l4: Number of neurons in the fourth layer.
• samples_prev: Total number of samples used for training.
• step: Iteration number at which the model stopped improving.
• epoch_cosine_similarity, epoch_loss, epoch_lr, epoch_mae, epoch_mape, epoch_mse: Metrics calculated at the end of each epoch to monitor the model’s performance during training.
• evaluation_cosine_similarity_vs_iterations, evaluation_loss_vs_iterations, evaluation_mae_vs_iterations, evaluation_mape_vs_iterations, evaluation_mse_vs_iterations: Metrics for evaluating the model on the validation or test set after each iteration.
• loss_test, loss_train, loss_val: Metrics representing the final loss after training.

.py Files:\ "train.py": Python algorithm for training.

.ipynb Files:\ "resultsbestmMSE2.ipynb": Jupyter notebook for tests and validations.

"resultsbestmMSE.ipynb": Notebook for comparing metrics from the training process

.h5 Files:\ "best_model_(90, 90, 1, 1).h5": The model with the best metrics resulting from the training. The name indicates that the model contains 4 layers, with 90 units in the first two layers and 1 unit in the last two layers.

• Backend: The model was trained using TensorFlow.
• Keras Version: The model was trained using Keras version 2.10.0.
• Model Configuration :
  • The model architecture is defined as a Sequential model with the following layers:
    • InputLayer: Takes a sequence input with a shape of (20, 1).
    • LSTM (2 layers): Both LSTM layers have 90 units, tanh activation, and recurrent sigmoid activation.
    • GRU: Has 1 unit with tanh and sigmoid activations.
    • SimpleRNN: Contains 1 unit and tanh activation.
    • Dense: Contains 1 unit with relu activation.
• Training Configuration:
  • Loss Function: mean_squared_error (MSE)
  • Monitored Metrics:
    • MSE: Mean Squared Error.
    • MAE: Mean Absolute Error.
    • MAPE: Mean Absolute Percentage Error.
    • Cosine Similarity: Cosine similarity between predictions and actual values.
  • Optimizer: Adam with:
    • Learning rate: 0.0093
    • Tuning parameters: beta_1 = 0.9, beta_2 = 0.999, epsilon = 1e-07.

RAR File 2: IIR_Digital_Filtering.rar

Description:
This .rar file contains a MATLAB implementation of IIR digital filters (Infinite Impulse Response Filter), along with text files containing pH measurements. The goal is to apply the IIR filters to the pH signals to analyze the filtered response.

• .txt Files: "SignalWithoutPerturbations.txt","SignalWithPerturbation-Agitator.txt","SignalWithPerturbations-Hits.txt"Datasets with pH signals at various levels with and without perturbations.
• .m Files: "iir_filters.m": MATLAB algorithm for filtering pH signals.