Monitoring of cooling tower water pumps using Arduino data acquisition device

Investing equipment to monitor the condition of assets through vibration can be expensive. Recent development of alternative microcontrollers has enabled researchers to study its potential to replace expensive equipment for monitoring purposes. This study aimed to investigate the potential of alternative microcontrollers to be used as devices to monitor the condition of water pumps. An internet of things (IoT) device was developed that can measure vibration and be applied to monitor water pumps. The vibration data was obtained when the pumps were operated and sent to the condition-based monitoring system (CBMS) database via Wi-Fi network. The vibration data was observed, and current and future condition was identified through vibration analysis.


Introduction
A water pump is a rotating machine that uses blades or impellers to increase water flow to move it from one point to another.A centrifugal water pump is a rotating machine in which flow and pressure are generated dynamically.The impeller of a water pump is rotated by an electric motor to increase water flow as it enters axially and is expelled by centrifugal force along its circumference through the tips of the impellers.Water will gain both velocity and pressure as it flows through the impeller and radially outward into a diffuser or spiral chamber/enclosure from which it exits.
The longevity and performance of a water pump depends on its regular maintenance.Maintenance activities include regular inspection of the pump's components and replacing the pump parts that are worn or damaged which incurs a fixed maintenance cost.However, excessive usage of water pumps can also damage the equipment and significantly decrease the pump's performance.Any irregularities in the water pump's performance should be addressed so that the water pump can operate effectively and perform its intended function without experiencing any significant problems.Water pump failure can result from bearing failure, misalignment, undesired physical load, aging components, and wear and tear.[1,2] which can be detected through emitted vibration [1].Therefore, it is important to understand the water pump current conditions and predict the point of failure in order to reduce any risk of affecting production.
In preventive maintenance, equipment is routinely monitored and scheduled maintenance activities are carried out to prevent unexpected failures in the future.Scheduled maintenance is performed with the goal of restoring the equipment to its original condition so that it runs efficiently and extends its life [3,4,5].Condition-based maintenance approach recommends maintenance decisions based on the information collected through condition monitoring [6].Based on feedback from certain indicators, maintenance is performed when a certain threshold is reached.As preventive maintenance does not consider the current health of the machine, unavoidable and unplanned machine failure is still likely, resulting in unnecessary costs [7].Manufacturers also often emphasize accelerating production output, sacrificing and delaying recommended maintenance downtime to generate more revenue [8].
Recent development of artificial intelligence and Internet of Things (IoT) has enabled researchers to apply and identify trends and anomalies in machine operation, and then predict potential failures so manufacturers can act before it's too late.Predictive maintenance is gaining interest because it is a form of real-time monitoring technology that is one of the foundations of Industrial Revolution 4.0 [9].With the advancement of wireless sensor networks, it is possible to use a low-cost IoT device to monitor machinery and equipment [10].The collected data, such as vibrations which usually in terms of vibration magnitude as in the ISO10816, is sent to the cloud for storage and further processing using IoT protocols and technologies [11].The availability of historical data enables analysis to be performed and future failures of machine equipment to be predicted, even if data from previous failures is not available [12].Vibrations are measured using sensors that collect vibration and temperature data over a period and transmit the information via the IoT to user interface software for display and simultaneous storage of the data in a database or cloud.The vibration of the water pump's electric motor, caused by various reasons, is important and requires constant monitoring to avoid serious faults.
One of the most common causes of water pump failure is cavitation.It is a symptom of insufficient net positive suction head when the absolute pressure of the liquid at the impeller inlet approaches the vapor pressure of the liquid [13].Another common cause of water pump failure is pulsation of the pump flow.It occurs when the water pump is operating near its shutoff level.The gauges on the pump's discharge line then fluctuate.When the water pump stops, there is a rapid volume shift that pulls water back into the pump, causing the pressure to rise.An imbalance in the pump impeller also has a strong effect on the efficiency of the pump and the life of the bearings.Imbalance occurs if the principal axis of inertia of the rotor is not coincident with its geometry axis [14].Imbalance in the pump impeller can lead to overheating, shaft deflection, bearing failure, excessive vibration, mechanical seal or packing failure, or complete pump failure.A bent shaft also contributed to the failure of the water pump.Mechanical overload due to damage during erection or improper handling, shock during operation or misalignment of the water pump, internal stress due to vibration during shipping, improper material handling or elevated temperatures during operation, or stacking stress during assembly due to severe shrink joints with mating parts such as a turbine wheel can result in a bent shaft, one of the most common problems in water pump failures [14].
Alternative microcontrollers have limitations in terms of sampling rates, range, storage, precision, and accuracy.It is not clear, although it has been previously studied those alternative microcontrollers have potential of being device for condition-based monitoring, whether alternative microcontrollers are able to detect failures of water pumps.This study aimed at investigating the potential use of alternative IoT microcontroller to measure, monitor and predict vibration emitted by water pumps.Arduino microcontroller, specifically Arduino Mega, has been employed which was previously tested by previous studies of its potential.

Methodology
This study used Arduino microcontroller data acquisition (ADaq) and inexpensive accelerometers developed to capture vibration data from running water pumps in a manufacturing factory's cooling tower.

Development of Arduino Data Acquisition
The ADaq was developed based on previous studies [15], consisting of components as shown in Table 1.The Arduino Mega was connected to WiFi shield which also has an SD Card reader to store raw acceleration data.The accelerometer was applied with magnetic strip to easily attached to the pump.Due to the limitation of the microcontroller, the vibration was measured at a sampling rate of 537 samples per second which enabled the system to detect any vibration energy in the frequencies up to about 250 Hz.The ADaq was working standalone, no computer was connected to it.The data it measured was sent to cloud system, which was later retrieved for analysis.The raw data was also measured and stored in the SD Card for later analysis.Due to dynamic memory limitations, the ADaq can only store and measure 1s at a time before the storage needs to be cleaned to prepare for the next measurement.The sensor was calibrated automatically via gravity.

Measurement Setup
This study employed a cooling tower water pump as shown in Figure 1.The water pump motor is a category of rotodynamic pumps for industrial applications, having the power rated at only 7.5 kW with operating speed of about 1500 rpm.Following the International Standard ISO 10816, the vibration magnitude that can be considered safe and normal for newly commissioned machines falls under 3.2 mm/s.The monitoring method of water pump's condition is based on the extraction of the vibration data captured using an ADXL345 accelerometer sensor.The data collection is recorded in real time as sets of accelerations consisting of x-axis, y-axis, and z-axis, subsequently converted into root-mean-square (RMS).The sensors were mounted on the side of the induction motor of the water pumps as shown in Figure 1 with its x-axis direction parallel with the shaft axial direction, while y-axis is in horizontal direction (leftright) and z-axis is in vertical direction(up-down).

Analysis of Recorded Data
The vibration was measured and subsequently converted to root-mean-square (RMS) before sending it to the cloud system.The raw data was later plotted as in graph (refer section 3) and later Fast Fourier Transform (FFT) analysis was conducted.Linear regression analysis can be used to predict the link between the vibration measured over time.The equation used in linear regression was  = 0 + 1 where 0 was the intercept of a line, 1 was linear regression coefficient,  was the predicted vibration and  was number of measurements over a time.The whole process for the analysis of linear regression was done using MATLAB software.There was a short period of time where high vibration magnitude can be observed (as observed in Figure 3).This, however, was normal and due to the pump starting process.

Time domain and frequency domain vibration of the water pump
A further analysis using time domain and frequency domain for water pump was carried out.The water pump was previously mentioned running at a constant 1465 RPM.The vibration measured a maximum of about 2 m/s 2 and a minimum of about -1.5 m/s 2 as shown in Figure 4.This was a single data measured at the water pump which was about 1 second with sampling frequency of 537 Hz.  Figure 5 shows the frequency domain of data measured at the water pump.A significant amplitude of vibration in the 1x RPM 5x RPM and about 9x RPM is observed, at the amplitude of 0.47 m/s 2 , 0.32 m/s 2 and 0.35 m/s 2 .
Figure 5 FFT Analysis of the vibration measured at the water pump.

Predicting conditions of water pump using linear regression
From the MATLAB using fitlm, the following regression model equation is obtained for forecasting (Figure 6). = 0.689 + 3.7 − 7………. (1) The fitted model has a positive gradient which can be used to predict future vibration emission.If warning level is identified, an alarm can be set to inform the maintenance worker.

Conclusion
The IoT device setup using Arduino Mega 2560, Cytron ESP8266 wifi shield and AXDL345 accelerometer sensor has been developed and constructed for this study.The device was able to capture vibration conditions that were located in an induction motor of the water pump and sent to the cloud.
The measured data was extracted and analyzed.It showed the condition of the water pump for 9 days and further analysis in time and frequency domain were achieved.This study showed that alternative microcontrollers have potential to be used for condition monitoring of water pumps.In addition, the study was completed using only one accelerometer sensor mounted on the water pump, suggesting that with only one accelerometer it is sufficient to setup a predictive maintenance system with the scope of monitoring current trend, predicting future trend and even fault diagnosis activity.

Figure 1 3 .
Figure 1 Left: The cooling tower water pump.Right: IoT Device Installation in Cooling Tower Water Pumps

Figure 2
Figure2shows an example of vibration measured at the water pump for one day.It can be seen that the vibration measured at the water pump does not fluctuate much across the day.The x-direction vibration has the greatest value and is followed by the y-direction vibration and the z direction vibration.

Figure 2
Figure 2 Daily vibration measured at the water pump.

Figure 3
Figure3shows the r.m.s.acceleration measured at the water pump.It can be observed that the vibration varied about 0.2 m/s 2 r.m.s.acceleration.Vibration remained almost constant throughout the 9 days of measurement.There was a short period of time where high vibration magnitude can be observed (as observed in Figure3).This, however, was normal and due to the pump starting process.

Figure 3
Figure 3 Nine days r.m.s.acceleration measured at the water pump.

Figure 4
Figure 4 Time domain of data measured at the water pump.

Figure 6
Figure 6 Fitted linear regression for data of the water pump.

Table 1
Arduino Data Acquisition System Components