Key points
Maintenance based condition monitoring of centrifugal pumps is critical for many industrial processes. Owing to the need for delivering advanced products with complex requirements while adhering to tight schedules, there is a growing demand for improved machine lifetimes from industry and production organisations.
Under these conditions, the damage to machines incurred by continued use must be balanced against the need for continuous output. Frequent machine failures can result in costly repair costs or, if ignored, can lead to catastrophic problems, production interruptions and supply failures.
The reliance on complex machines means that profitability would be impacted by any failures. This risk exists for many reasons, such as the lack of availability, the cost of spares, the cost of labour breakdown, the cost of secondary damage, and the risk of human and environmental injury.
Therefore, there is an essential need to utilise a smart diagnosing system based on Artificial Intelligence (AI) along with other advanced tools such as the Internet of Things (IoT).
Centrifugal pumps (operation and maintenance)
Pumps are typically available in various shapes and sizes, such as: centrifugal, propeller and positive displacement, turbo. There are various reasons that cause the pumps to vibrate, and it is possible to consider the vibration frequencies to distinguish basic factors that could be hydraulic or mechanical.
The emphasis in this article is put on the centrifugal pumps, in which the centrifugal pumps are vital rotating machines, where failure-free operation is necessary for output. Also, the centrifugal pump is one of the most common rotating machines.
Many different potential faults can be occurring to a centrifugal pump, many of which can be categorised as hydraulic and mechanical faults. Hydraulic defects include cavitation, hammering of water, and turbulence.
Mechanical defects include misalignment, imbalance, impeller damage as shown in Figure (1), mechanical looseness as shown in Figure (2), cracked bearing and broken seal, and bearing damages, where Figure (3) shows a ball bearing with a damaged inner race.
Utilisation of AI in the condition monitoring of centrifugal pumps
There have been tremendous research works have been implemented on the application of AI in the condition monitoring of rotating machines including the centrifugal pumps.
One of these research works has been conducted on the centrifugal pumps, where a centrifugal pump experimental setup that simulates the industrial pumps has been used in this research, and seven different conditions were tested (normal, imbalance, misalignment, impeller damage, bearing damage, mechanical looseness, and cavitation).
An accelerometer connected to a Data Acquisition device was used to acquire the vibrational data/signals from the centrifugal pump. LabVIEW software was used to capture the signals and visualise them in both time and frequency domains, where the signals can be also saved and taken as raw data for the next step of pre-processing. Figure (4) shows the centrifugal pump experimental setup.
Using Internet of Things (IoT) with Artificial Intelligence for Centrifugal Pump Condition Monitoring
Based on the previous and recent research works on the centrifugal pumps condition monitoring using AI along with the optimisation and pre-processing algorithms such like Genetic Algorithm and Wavelet Transform methods respectively, adding and integrating IoT into this loop to have a powerful diagnostic system for the rotating machines and most effectively for the centrifugal pumps would be a significant leap in the field of machine condition monitoring.
Therefore, considering the previous successful utilisation of AI in the condition monitoring of the centrifugal pumps can be further enhanced by taking into account an advanced diagnostic system condition monitoring based AI-IoT.
IoT can be utilised in the field of condition monitoring and particularly for the centrifugal pumps through the integration with AI, where the smart sensors that are mounted on the pumps can transmit the acquired data to AI system for the processing and classification of the faults.
The sensors including the accelerometers for the vibration parameter can acquire numerous data per time, and that is what can be considered as big data.
Moreover, these sensors are categorised as smart ones in which the transmission of huge and big data is ensured to be fast and secured. The concept of how IoT can be integrated into AI to diagnose the condition of centrifugal pumps is depicted in Figure (5).
From Figure (5), there are three main stages for the automatic diagnostic system based IoT and AI, and the three stages are: (1) Extraction of the data from a centrifugal pump (there can be many centrifugal pumps and other rotating machines) using smart sensors which are transmitting the data to the cloud storages and servers; (2) the data will be moving through an acquisition and a signal processing system in which the amplification and filtration are implemented, and the data will be captured and visualised in its original form (time-domain); (3) the acquired data will be pre-processed to consider the best selection of the data, and the selected data (features) will be forwarded to the AI system in which the classification of the conditions will be implemented based on the applied AI methods and algorithms.
Artificial Intelligence
Artificial intelligence (AI) can be used in automated fault detection methods to simulate mental capabilities with the help of computer systems. There are some parallels between the biological neuron and the artificial neuron, as the latter was inspired by the former biological neuron as shown in Figure 3, which it consists of main components such as the main body (cellular body), dendrite and axon.
Dendrite's function is to collect the associated neurons' signals, and the neuron receives the electrochemical signals from various sources, and the signals are then sent when it gets exciting (forming spikes). There are a sufficient number of neurons in synapses that are bound together. On the other hand, the artificial neuron has almost the same mechanism of the biological one.
One of the popular artificial intelligence types is the Multilayer Perceptron (MLP) which consists of three layers, namely, input, hidden, and output layer of neurons, and there can be many hidden layers between the input and output layers.
The input layer is responsible of receiving the input features from the external environment (i.e. like machines), the hidden layer plays a role of processing the inputs and mapping the input layer to the output layer, and the output layer works to classifying and recognising the different conditions.
Pre-processing methods
Pre-processing is an important element in condition monitoring based AI, where the acquired and extracted features (data) have to be enhanced through the pre-processing methods to obtain good features that would be forwarded then to the AI system.
There are various methods can be utilised for the pre-processing of the signals or data such as frequency-domain, and time-frequency domain methods, viz. the Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), Discrete Wavelet Transform (DWT), and Wavelet Packet Transform (WPT). The Time-Frequency method has a better ability in handling the non-stationary signals in which provides more useful features for the AI.
Insight into the next step
Most recently, AI has been effectively applied for the automatic fault diagnosis of the rotating machines including the centrifugal pumps. However, the integration of IoT into AI is still at its early stages, as there are a few attempts and approaches of such combination in the field of machine condition monitoring. It is extremely a big enhancement for the AI-based condition monitoring, and particularly for the critical machines like the centrifugal pumps.
By integrating IoT into AI, many and even numerous centrifugal pumps can be remotely monitored as they are connected to different smart sensors that sense out many parameters like the vibration, temperature, pressure, and electrical signals.
IoT allows these sensors to instantly and securely transmit the big data from different machines and to be online connected to the AI systems through a network. Introducing such a smart diagnostic system in industries and particularly in the maintenance of machines would be a good idea that any condition monitoring engineer can adopt.
