Sensor-Based Monitoring Data from an Industrial System of Centrifugal Pumps

Abstract

We present a detailed dataset collected via a wireless IoT sensor network monitoring three industrial centrifugal pumps (units A, B, and C) at the Italian Aerospace Research Centre (CIRA), along with the methods for data collection and structuring. Background: Centrifugal pumps are critical in industrial plants, and monitoring their condition is essential to ensure reliability, safety, and efficiency. High-quality operational data under normal operating conditions are fundamental for developing effective maintenance strategies and diagnostic models. Methods: Data were gathered by means of smart sensors measuring motor and pump vibrations, temperatures, outlet fluid pressures, and environmental conditions. Data were transmitted over a WirelessHART mesh network and acquired through an IoT architecture. Results: The dataset consists of eight CSV files, each representing a specific pump during a distinct operational day. Each file includes timestamped measurements of displacement, peak vibration values, sensor temperatures, fluid pressure, ambient temperature, and atmospheric pressure. Conclusions: This dataset supports advanced methodologies in feature extraction, multivariate signal analysis, unsupervised pattern discovery, vibration analysis, and the development of digital twins and soft sensing models for predictive maintenance optimization.

Dataset: Data are available at https://doi.org/10.5281/zenodo.15301820 (accessed on 13 June 2025).

Dataset License: CC-BY 4.0

1. Introduction

1.1. Problem Statement

In the industrial sector, the most commonly adopted maintenance strategies are currently Corrective and Preventive Maintenance. However, both approaches can result in significant costs for companies. Corrective Maintenance (CM) may lead to unexpected machine downtime, while Preventive Maintenance (PM) often involves replacing components that may still be fully functional. In contrast, implementing an effective Predictive Maintenance (PdM) plan, intervening only when and where necessary, can lead to substantial cost savings and improved overall operational efficiency across the industrial organization.

In [1] PdM is defined as “a philosophy or attitude that, simply stated, uses the actual operating condition of plant equipment and systems to optimize total plant operation. A comprehensive predictive maintenance management program uses the most cost-effective tools (e.g., vibration monitoring, thermography, tribology) to obtain the actual operating condition of critical plant systems and based on this actual data schedules all maintenance activities on an as-needed basis”. Therefore PdM aims to identify trends, anomalies, and degradation in the initial stages, so that the necessary ones can be implemented countermeasures. PdM is implemented by identifying one or more parameters that are measured and then analyzed using appropriate mathematical models to estimate the remaining useful life (RUL) of a system before failure occurs. Various techniques are employed for this purpose, including vibration analysis, thermography, current consumption analysis, detection of abnormal vibrations, and more. Deviations in measured values from the normal operating condition indicate progressive degradation and, ultimately, enable the prediction of failure timing. The application of PdM is now largely enabled by advancements in IoT. Real-time, connected IoT solutions enable companies to collect large volumes of operational data from equipment. This wealth of information allows for the implementation of algorithms capable of reliably predicting failures and identifying early signs of degradation in production processes. Hence, the availability of machine operational datasets is a critical enabler for the effective implementation of a PdM paradigm.

1.2. Related Works

In both the academic literature and online resources, several sets of machinery operational data are available to support preliminary analyses for PdM. The following datasets have currently been identified:

• NASA Bearing Dataset [2]: each of the three datasets made available is composed of individual files, each representing a one-second snapshot of the vibration signal recorded at specific time intervals. Every file contains 20,480 data points, with a sampling rate of 20 kHz. The file name encodes the timestamp indicating when the data was collected. Each row in a data file corresponds to a single data point. Larger gaps between timestamps (as reflected in the file names) indicate that the experiment was resumed on a subsequent working day.
• MFPT Bearing Fault Dataset [3]: the MFPT dataset includes 23 distinct datasets collected from machinery operating under various fault conditions. The first 20 datasets were recorded on a bearing test rig: 3 represent bearings in good condition, 3 feature outer ring defects under constant load, 7 contain outer ring defects under varying loads, and 7 include inner ring defects under different load conditions. The remaining three datasets were collected from real-world machines, specifically an oil pump bearing, an intermediate speed bearing, and a planetary bearing. Each dataset contains an acceleration signal “gs”, the sampling frequency “sr”, the shaft speed “rate”, the weight of the load “load” and four critical frequencies that identify the different locations of a possible failure: the Ball Passing Frequency on the Outer ring (BPFO), the Ball Passing Frequency on the Inner ring (BPFI), the Fundamental Train Frequency (FTF), also known as cage frequency, and the Ball Spin Frequency (BSF).
• UCI Accelerometer Dataset [4]: this dataset contains vibration data from a cooling fan equipped with weighted blades, recorded using accelerometers. It is suitable for tasks such as prediction, classification, and other applications involving vibration analysis, particularly in engine-related systems.

1.3. About This Paper

The rest of this paper is organized as follows: Section 2 provides a detailed description of the centrifugal pump dataset made available in this study. Section 3 presents the methods used to create the dataset, including the deployment of a wireless IoT sensor network and the development of a SW data platform for data collection, storage, and visualization. Section 4 describes the preliminary data understanding phase, conducted to assess the consistency, variability, and distribution of sensor measurements. Section 5 explores various application scenarios aimed at supporting the development of advanced methodologies for industrial monitoring, maintenance, reliability analysis, and intelligent diagnostics. Finally, Section 6 concludes the paper with reflections on the implications of the study and future research directions.

2. Data Description

Centrifugal Pump Dataset

This dataset consists of eight CSV files, each containing data collected from a specific centrifugal pump during a different operational day. Currently, recordings from 10 April, 11 June, and 30 October 2024 are available. The filename uniquely identifies both the pump unit (A, B, or C) and the corresponding day of operation. As of the time of writing, data has been collected from three different operational days, each with its own time duration.

Table 1. Dataset structure with operating ranges of centrifugal pumps for each operational day.

Filename	Centrifugal Pump Unit	Operational Day	Operating Range
			Startup Timestamp	Shuthown Timestamp
A_2024-04-10.csv	A	10 April 2024	12:28:30	12:49:28
B_2024-04-10.csv	B		12:51:42	12:56:41
C_2024-04-10.csv	C		12:58:16	13:10:14
A_2024-06-11.csv	A	11 June 2024	10:00:25	11:21:24
B_2024-06-11.csv	B		07:08:08	13:07:33
C_2024-06-11.csv	C		10:52:55	11:19:54
A_2024-10-30.csv	A	30 October 2024 *	10:59:15	13:19:13
B_2024-10-30.csv	B		08:28:33	11:05:56

* Data for centrifugal pump C are not available for this day, as the pump was turned off.

provides the dataset structure along with a summary of the operating ranges, defined by the startup and shutdown timestamps, of all centrifugal pumps, grouped by operational day.

Each CSV file contains eleven columns: one for the timestamp and ten for measured values. The column headers are those listed in

Table 2. CSV header column names, their descriptions, and UoM.

Column Name	Description	UoM
Timestamp	temporal timestamp indicating when the measurement was taken, formatted as YYYY-MM-DD hh:mm:ss	–
X_ACR_Mot.PV	vibrational velocity measured by the motor accelerometer	m/s
X_ACR_Mot.SV	peak value measured by the motor accelerometer	m/s2
X_ACR_Mot.TV	contact temperature of the motor accelerometer	°C
X_ACR_Pmp.PV	vibrational velocity measured by the pump accelerometer	m/s
X_ACR_Pmp.SV	peak value measured by the pump accelerometer	m/s2
X_ACR_Pmp.TV	contact temperature of the pump accelerometer	°C
X_Temp.SV	motor casing temperature	°C
X_Pres.SV	outlet fluid pressure from the pump	bar
Barometer	atmospheric pressure	mbar
Temperature	ambient temperature	°C

, where the prefix X represents the centrifugal pump unit label (e.g., A, B, or C).

The method employed to determine the peak value warrants some clarification. This value is calculated internally by the accelerometer as the maximum acceleration detected within the one-second interval preceding each measurement. During this interval, the acceleration signal is filtered using a high-pass filter with a cutoff frequency of 1 kHz. According to findings reported in the literature [5], high peak values in the high-frequency range, typically between 1 kHz and 20 kHz, are indicative of defects in motor and pump bearings and gears.

Our dataset differs from those reported in Section 1.2 by providing additional operating parameters of centrifugal pumps beyond vibrational data, specifically the temperatures of both the motor and the pump, as well as the outlet fluid pressure. Furthermore, it includes the simultaneous monitoring of environmental conditions through measurements of ambient temperature and atmospheric pressure. In addition, it enables investigations into the mutual interactions between the centrifugal pumps during their simultaneous operation.

3. Materials and Methods

This section presents a detailed description of the methods used to create the final centrifugal pump dataset provided in this study. The dataset was collected through an IoT wireless sensor network and stored on a data monitoring platform, the architecture of which is described in the following sections.

3.1. Target Plant and IoT Data Acquisition System

The target plant consists of a group of three industrial centrifugal pumps that supply demineralized water to the boilers of a heating plant. This plant produces the steam required by the research test facilities at the CIRA (Italian Aerospace Research Centre). Each centrifugal pump, labeled as A, B, or C, includes an electric motor coupled with a mechanical multistage pump. The multistage design increases the fluid’s pressure progressively at each stage, ensuring efficient water transfer.

Table 3. Nameplate data of motor and pump.

Electric Motor
Power	Shaft Speed	Voltage	Poles	Frame Size (H)
110 kW	2980 rpm	400 V	2	315 mm
Mechanical Multistage Pump
Power	Flow Rate	Pump Head	Max Pressure	N° of Impellers
110 kW	45 m3/h	450 m	40 bar	7

summarizes the nameplate specifications of both the electric motor and the mechanical multistage pump.

Furthermore, each centrifugal pump is equipped with four smart wireless sensors to monitor operational parameters:

• Two accelerometers: one mounted on the pump and the other on the motor, used to monitor vibration levels, both of which are also equipped with a temperature sensing element for contact temperature measurement.
• One temperature sensor: positioned on the motor casing to track thermal conditions.
• One pressure sensor: used to measure the flow rate of the outgoing fluid.

Moreover, two additional external wireless sensors are added to monitor ambient temperature and pressure to set up a small-scale environmental monitoring station. All sensors perform measurements at a sampling frequency of 1 Hz.

Sensors are integrated into a wireless mesh network based on the industrial WirelessHART protocol. This network communicates with a wireless gateway, which serves as the data acquisition unit. The gateway collects the sensor data and timestamps them in UTC with a resolution of one second. The data are subsequently retrieved by a centralized IoT Data Monitoring platform, as shown in Figure 1.

3.2. Iot Data Monitoring Platform

The SW architecture of the IoT Data Monitoring platform is designed to collect, store, and display data from multiple sensors, each potentially using different data formats. It is built using three open-source components (Telegraf, InfluxDB, and Grafana), which together form what is commonly known in the literature as the TIG stack. Figure 2 illustrates the system’s architecture and data flow. This setup manages the flow of data from IoT sensors (more precisely, from the gateway that aggregates data from sensors connected via the WirelessHART protocol) and enables users to visualize the data through a web browser.

The first component of the architecture is responsible for collecting all the measurements gathered by the sensors in the monitoring system. This is achieved using an SW agent called Telegraf [6], which is specifically designed for data collection. Telegraf supports the collection and transmission of a wide variety of data types, including database systems, SW platforms, and IoT sensors (using popular protocols such as MQTT, ModBus TCP, and OPC-UA).

In our implementation, data will be collected by Telegraf using the ModBus TCP protocol [7] from a specific IP address assigned to the gateway. The collected data will then be sent to a TSDB. TSDBs are specifically designed to handle time series data, which consists of arrays of numbers indexed by time (either as a specific date or within a range of dates). These databases support graphical queries, allowing users to visualize streaming data easily. Additionally, TSDBs provide native support for basic calculations, such as multiplication, addition, and the ability to join multiple time series into a new historical series. Users can also filter data based on temporal patterns, such as the day of the week or the time of day.

In our SW architecture, InfluxDB [8] was chosen as the TSDB. It is a high-performance, open-source TSDB designed to handle high write and query loads. InfluxDB also includes a built-in, web-oriented GUI that allows users to view the graphs of sensor data in near real-time. These graphs can be dynamically and interactively updated based on user requests. Time series data are typically visualized using line plots, and InfluxDB’s graphical interface enables users to select one or more metrics and/or tags to generate graphs and statistics. Alternatively, more advanced open-source graphical editors are available, which can interface with TSDBs, typically using HTTP and/or REST protocols, and offer monitoring frameworks, dashboards, statistical analysis tools, and automation features.

One such example is the widely used Grafana UI framework [9], which was employed in our implementation. Grafana allows users to visually represent time series data stored in InfluxDB, offering a variety of chart types (e.g., bar charts, line charts, etc.). It queries the InfluxDB TSDB via its REST API. A key feature of Grafana is its ability to aggregate multiple data sources into a single dashboard, incorporating panels arranged in rows. Additionally, users can create custom dashboards and share them with others.

Figure 3 presents the time evolution of the accelerometer-related variables for centrifugal pump A, as visualized through Grafana UI.

Summarizing, the IoT Data Monitoring platform follows a structured workflow to collect, process, and visualize data from industrial sensors. The process can be divided in four main stages:

• Data Acquisition (WirelessHART Gateway): sensors installed on equipment continuously measure parameters and transmit data, via the WirelessHART protocol, to a WirelessHART Gateway that acts as a local data aggregator.
• Data Collection (Telegraf): Telegraf handles the extraction of data from the WirelessHART Gateway and prepares them for storage.
• Data Storage (InfluxDB): the processed data from Telegraf is stored in InfluxDB, which ensures efficient storage, quick retrieval, and enables historical analysis of the data.
• Data Visualization (Grafana): Grafana is connected to InfluxDB to retrieve stored data and present it in the form of interactive dashboards, charts, and graphs. Users can access these visualizations via a standard web browser, allowing real-time monitoring as well as historical trend analysis.

4. Results

A preliminary data understanding phase was carried out to assess the consistency, variability, and distribution of the sensor measurements. Through visual inspections and descriptive statistics, distinct dynamic behaviors of the data were identified. For example, temperature and pressure values evolved smoothly over time, displaying relatively low variability and near-symmetric distributions. In contrast, vibration signals, particularly the peak values, exhibited sharp fluctuations, long-tailed distributions, and pronounced skewness (

Table 4. Descriptive statistics for centrifugal pumps A, B, and C on 10 April 2024.

Variable	Mean	SD	Median	Trimmed	Min	Max	Skew	Kurtosis
A_ACR_Mot.PV	0.002	0.000	0.002	0.002	0.001	0.003	1.129	1.752
A_ACR_Mot.SV	1.708	8.387	0.427	0.438	0.388	456.626	37.005	1843.644
A_ACR_Mot.TV	28.284	3.431	29.063	29.106	17.977	31.367	−2.161	3.570
A_ACR_Pmp.PV	0.001	0.001	0.001	0.001	0.000	0.004	2.960	7.045
A_ACR_Pmp.SV	2.761	13.663	0.526	0.533	0.445	291.623	17.041	344.468
A_ACR_Pmp.TV	23.256	2.131	22.625	23.419	18.219	26.578	−0.480	0.172
A_Pres.PV	3.754	11.162	0.452	0.452	0.440	41.978	3.107	7.665
A_Temp.PV	26.998	3.078	27.241	27.537	17.929	31.083	−1.615	2.550
B_ACR_Mot.PV	0.001	0.000	0.001	0.001	0.000	0.002	1.183	1.001
B_ACR_Mot.SV	1.662	4.282	0.506	0.506	0.428	20.440	3.475	10.201
B_ACR_Mot.TV	19.166	0.327	19.305	19.224	18.203	19.422	−1.314	0.668
B_ACR_Pmp.PV	0.001	0.001	0.001	0.001	0.000	0.004	2.876	7.799
B_ACR_Pmp.SV	1.970	5.582	0.484	0.495	0.000	26.223	3.545	10.670
B_ACR_Pmp.TV	20.966	1.264	20.961	21.030	18.313	22.852	−0.557	−0.503
B_Pres.PV	3.057	10.156	0.478	0.479	0.464	43.211	3.694	11.647
B_Temp.PV	19.175	0.590	19.046	19.094	18.210	21.575	2.028	5.436
C_ACR_Mot.PV	0.001	0.001	0.001	0.001	0.001	0.003	1.606	1.716
C_ACR_Mot.SV	2.724	5.707	0.423	1.143	0.384	31.866	2.290	3.933
C_ACR_Mot.TV	22.318	2.430	23.203	22.406	18.789	24.859	−0.199	−1.764
C_ACR_Pmp.PV	0.001	0.001	0.001	0.001	0.000	0.004	1.984	2.107
C_ACR_Pmp.SV	4.484	9.938	0.413	1.815	0.376	37.670	2.084	2.460
C_ACR_Pmp.TV	21.546	1.738	22.547	21.678	18.656	23.141	−0.546	−1.522
C_Pres.PV	6.636	15.053	0.452	2.836	0.442	43.301	2.022	2.090
C_Temp.PV	23.605	3.178	25.448	23.616	18.822	28.557	−0.331	−1.470
Barometer	1018.078	0.059	1018.075	1018.076	1017.883	1018.271	0.356	0.276
Temperature	19.907	0.340	19.951	19.909	19.171	20.771	−0.074	0.071

,

Table 5. Descriptive statistics for centrifugal pumps A, B, and C on 11 June 2024.

Variable	Mean	SD	Median	Trimmed	Min	Max	Skew	Kurtosis
A_ACR_Mot.PV	0.003	0.000	0.003	0.003	0.001	0.003	−2.202	6.262
A_ACR_Mot.SV	13.160	3.177	13.734	13.719	0.543	17.497	−3.203	10.215
A_ACR_Mot.TV	35.108	5.727	36.500	35.715	23.766	41.477	−0.843	−0.499
A_ACR_Pmp.PV	0.004	0.001	0.004	0.004	0.001	0.004	−2.546	4.721
A_ACR_Pmp.SV	31.127	10.722	34.089	33.670	0.567	60.082	−2.146	3.770
A_ACR_Pmp.TV	36.483	3.860	37.852	37.243	27.148	39.781	−1.493	0.936
A_Pres.PV	41.597	0.659	41.534	41.571	40.406	43.129	0.315	−0.573
A_Temp.PV	35.347	4.645	36.736	36.051	23.963	40.506	−1.126	0.226
B_ACR_Mot.PV	0.001	0.001	0.001	0.000	0.001	0.002	0.469	0.192
B_ACR_Mot.SV	28.054	29.239	29.103	3.463	0.506	55.714	−1.979	5.702
B_ACR_Mot.TV	35.536	37.570	36.402	3.915	22.609	40.648	−1.121	−0.042
B_ACR_Pmp.PV	0.003	0.003	0.003	0.000	0.000	0.004	−2.130	3.660
B_ACR_Pmp.SV	23.710	25.820	25.240	3.828	0.000	70.682	−1.230	3.185
B_ACR_Pmp.TV	39.707	42.227	40.919	1.992	22.680	44.398	−1.803	2.314
B_Pres.PV	42.792	42.750	42.837	1.457	5.267	46.709	−2.342	51.374
B_Temp.PV	33.992	35.565	34.644	2.912	22.673	38.539	−1.038	−0.157
C_ACR_Mot.PV	0.001	0.000	0.001	0.001	0.001	0.002	3.532	10.481
C_ACR_Mot.SV	1.690	4.667	0.461	0.461	0.461	19.412	3.532	10.481
C_ACR_Mot.TV	24.907	1.397	24.539	24.539	24.539	30.211	3.532	10.481
C_ACR_Pmp.PV	0.001	0.000	0.001	0.001	0.001	0.003	5.812	31.803
C_ACR_Pmp.SV	1.979	8.805	0.526	0.526	0.526	55.285	5.885	32.658
C_ACR_Pmp.TV	24.795	1.931	24.477	24.477	24.477	36.484	5.885	32.658
C_Pres.PV	41.599	0.536	41.521	41.587	40.363	42.824	0.163	0.566
C_Temp.PV	29.487	2.974	30.207	29.584	24.479	33.381	−0.259	−1.418
Barometer	1011.880	0.213	1011.901	1011.899	1011.346	1012.268	−0.574	−0.414
Temperature	26.141	0.281	26.212	26.153	25.692	26.496	−0.388	−1.349

and

Table 6. Descriptive statistics for centrifugal pumps A and B on 30 October 2024.

Variable	Mean	SD	Median	Trimmed	Min	Max	Skew	Kurtosis
A_ACR_Mot.PV	0.003	0.000	0.003	0.003	0.001	0.004	−1.584	6.057
A_ACR_Mot.SV	16.066	3.759	16.061	16.315	0.466	24.247	−1.776	6.117
A_ACR_Mot.TV	39.961	7.256	42.633	40.795	21.211	50.273	−0.847	−0.295
A_ACR_Pmp.PV	0.004	0.001	0.004	0.004	0.001	0.004	−4.471	20.303
A_ACR_Pmp.SV	32.674	6.735	33.442	33.452	0.526	42.875	−3.042	14.423
A_ACR_Pmp.TV	37.634	2.660	38.547	38.075	28.961	41.297	−1.309	1.049
A_Pres.PV	38.792	6.418	39.884	39.821	0.670	42.487	−5.665	30.727
A_Temp.PV	38.973	5.536	40.633	39.611	23.660	49.677	−0.830	−0.032
B_ACR_Mot.PV	0.002	0.000	0.002	0.002	0.001	0.002	0.163	−0.143
B_ACR_Mot.SV	126.457	67.397	126.340	127.987	0.467	227.879	−0.123	−1.462
B_ACR_Mot.TV	37.318	5.353	39.258	38.213	18.992	42.805	−1.258	0.698
B_ACR_Pmp.PV	0.004	0.000	0.004	0.004	0.001	0.007	1.860	35.530
B_ACR_Pmp.SV	23.894	38.672	20.454	20.541	0.524	637.108	13.570	193.602
B_ACR_Pmp.TV	40.014	3.518	41.023	40.628	23.234	44.234	−1.500	1.775
B_Pres.PV	39.960	2.332	40.311	40.203	0.706	43.101	−7.600	112.337
B_Temp.PV	35.405	4.521	36.882	36.093	21.807	40.916	−1.182	0.578
Barometer	1022.019	0.323	1022.119	1022.034	1021.407	1022.473	−0.488	−1.104
Temperature	21.787	0.986	21.943	21.842	19.565	23.243	−0.327	−0.967

). The asymmetry was also confirmed by comparing the mean and the trimmed mean, which appear very similar for stationary variables but show noticeable differences for dynamic ones, indicating the presence of heavy tails and potential transient events. These statistical insights reflect the heterogeneous nature of the monitored signals.

The results of this preliminary analysis underscore the dataset’s potential for use in multivariate signal processing, temporal dynamic analysis, and condition monitoring.

5. Discussion

The sets of data represent a useful tool for the development of advanced methodologies of industrial monitoring, maintenance, reliability and intelligent diagnostics. In particular, the data were acquired in a variety of typical operating regimes, thus capturing the variability and dynamics of the normal industrial operation of centrifugal pumps (or other similar rotating machinery). The integration of statistical tools and machine learning models on this type of data is essential to extract useful information, define operational parameters, quantify process variability, and support effective decisions in the field of maintenance, optimization, and safety of plants [10,11]. The dataset therefore offers a broad basis for the experimentation and validation of advanced condition monitoring systems, as well as for the development of new analytical applications in industrial contexts characterized by sensor-rich environments. Some of the possible uses of the data are outlined below.

5.1. Operating Condition Benchmarking

The data provide a reference of nominal operating behaviour for comparison with the monitored machinery. It is suitable for training learning algorithms (autoencoder, one-class SVM, clustering) to recognize the normal operating values of variables. Once the normal operating conditions are learned, these models can then be implemented for anomaly or novelty detection on other assimilable equipment, flagging deviations from the learned normal state as potential failures.

5.2. Feature Extraction and Preprocessing

Since the signals represent only normal operation, the dataset is well-suited for developing, testing, and validating feature extraction and data preprocessing pipelines (noise reduction, normalization, missing value management) in realistic industrial scenarios without the risk of including hidden failure events.

5.3. Integration and Multivariate Signal Analysis

It is possible to perform multivariate analysis of temporal causality by integrating the different types of data (displacements, temperatures, pressures) coming from several machines in operation at the same time. This allows for studying how the different physical quantities influence each other during the normal operation of the plant. It enables the identification of correlations between variables and the development of models that take into account the mutual interaction and the influence that there may be in the operation of the machinery for a more complete and robust characterization of the operating conditions of the system. Through these methods, it is also possible to assess the impact of external environmental variables on machinery operating conditions.

5.4. Unsupervised Pattern Discover

The data lend themselves to exploratory analysis through the use of unsupervised learning methods, such as clustering and dimensionality reduction, to identify different operating states, load conditions, or transient regimes within the spectrum of normal operation as well as to explore the structure and variability of the dataset. This type of analysis can also help identify the most informative variables for subsequent analysis.

5.5. Vibration Analysis for Bearing Monitoring

Vibration signals from both the pump and the motor accelerometers can be used to develop, train, and test models to assess bearing condition. These data enable time- and frequency-domain analysis, allowing researchers to extract characteristics and evaluate algorithms aimed at detecting bearing wear or changes in dynamic behavior under normal conditions.

5.6. Digital Twin and Soft Sensing

Data can be used as input for the development of digital twins or soft sensors, allowing for the estimation or virtual detection of variables not directly measured, the validation of simulation models, or the optimization of processes based on data analysis.

6. Conclusions

In this study, our objective is to describe and provide a dataset derived from the operation of machinery, specifically centrifugal pumps. The dataset, collected through a wireless IoT sensor network and stored on an in-house developed SW data monitoring platform, serves as a valuable resource for subsequent investigations and modeling. For example, it can be used with Data Science (DS) methodologies to support the development of a PdM strategy.

Future steps may include the recording of additional operational days and the inclusion of diverse conditions, potentially involving signs of degradation in pump and motor behaviour. Furthermore, acquiring data at higher sampling frequencies could enable more detailed analyses in the spectral domain.