Future Of Refractory Maintenance in the Era of Industry 4.0

In the last two and half centuries, the world has witnessed three industrial revolutions. First industrial revolutions started in the 18th century, used water & steam power to mechanize production. The second industrial revolution took place at the end of 19th and at the beginning of 20th century, introduced electrical power which enabled mass production. The invention of transistors in 1947 was the beginning of the third industrial revolution and it used electronics & information technology to automate production[1]. Each revolution has contributed in building the foundation of the previous one and brought advancement in the manufacturing process. It has radically impacted our working environment & daily lives. Now, we are at the cusp of fourth industrial revolutions also known as Industry 4.0.

Industry 4.0 is a developing concept of automation of conventional manufacturing & industrial practices using smart technology such as digital twins, Internet of Things (IOT),?Big Data analytics, Cloud computing, advanced robotics, Machine Learning & Artificial Intelligence. Large scale machine-to-machine communication and the internet of things are integrated for increased automation, improved communication and self-monitoring[2].

Introduction to Artificial intelligence & Machine Learning:

How does AI/ML work?

Machine learning is part of Artificial intelligence that enables computer systems to think in a similar way to how humans do; learning & improving upon past experience. It works by exploring data, identifying patterns and making predictions. A typical AI/ML model follow the following steps as depicted below fig1:

Today’s AI is denominated by data driven techniques. Our technology infrastructures?such as sensors, system logs, user logs, storage availability e.t.c have grown up multifold leads to availability & accessibility of the data. These data could be an image, sound, text, temperature, pressure, machine RPM, refractory lining thickness, campaign durations, chemical ingredients?etc. For example, a smart factory may produce various images of the product components, which are classified as normal or defective. In case of refractory maintenance, input data can be residual thickness of refractory which can be measured using a laser, production parameter such as tapping temperature, processing time, lancing time, blowing time, maintenance data such as amount of mix used during gunning, refractory lining such as type of material used (castable, bricks), lining design & material quality.?The next step is to label the data. For instance, images of components from the smart factory can be labeled as ‘normal’ or ‘defective’. Images of refractory lining of teeming ladle can be labeled as ‘normal’, ‘ wear out’, ‘critical’ based on the residual thickness. These data are first cleaned and prepared in such a way that it can be used to train machine learning models. According to a survey conducted by Forbes, data scientists spend 80% of their time on data preparation[3] and hence data collection, cleaning & organizing data are critical steps in building up AI/ML projects.

Once, data is collected and organized in a particular fashion, the next step is feature engineering. Feature engineering is the process of using domain knowledge to extract features from the raw data via data mining techniques. It is a very important step in AI/ML which determines the quality of AI/ML outcomes. It drives the AI/ML model performance and governs the ability of the model to generate meaningful insights and ultimately solve the business problem. All the three steps so far discussed are summarized in the below table (fig-2) in form of sample data for better understanding.

In the context of refractory maintenance, refractories properties such as cold crushing strength at various temperatures, apparent porosity, bulk density, refractoriness, hot modulus of rupture, thermal expansion of coefficient, workability, water addition percentage, Alkali resistance and?specific process parameters could be the features which have an impact on the refractory wear. The next step is to select the appropriate machine learning algorithms model such as Linear classification, Support Vector machines(SVM), Artificial Neural Network(ANN), Convolution Neural Network (CNN), logistic regression, K-mean, decision tree, random forest etc where features are used as input data. Once these data get processed through the AI/ML analytical software, it gives output such as a forecast of refractory balance lifetime without maintenance, maintenance proposal including method, schedule, required time etc and holistic understanding of refractory wear mechanism such as impact of blowing time on refractory wear.?In order to give a flavour of algorithm, Artificial Neural Network is explained in detail.?

Artificial Neural Network(ANN)

The human brain can be described as a biological neural network - an interconnected web of neurons transmitting elaborate patterns of electrical signals. Dendrites receive input signals and based on these inputs, fire an output signal via an axon. One of the key elements of a neural network is the ability to learn, adapt and change its internal structure based on the information flowing through it[4]. Inspired by our biological neural network, an artificial neural network is developed which is basically a mathematical model that uses learning algorithms to store the information. Since neural networks are used in machines, they are collectively called Artificial Neural Networks. Similar to our brain, neural networks are built up of many nods, representing neurons and many connections between them, representing axons and dendrites which carry information. These connections are weighted.

The simplest form of Neural network is perceptron. A perceptron follows the feed-forward model meaning inputs are sent into the neuron which is processed and gives output. The perceptron algorithm calculate the sum of each input multiplied by its weight and passed through activation function and it is represented as y(x) = f(∑ni=1WiXi). Single Neuron with activation function is represented below in fig-3:

Single Layer Perceptron : One input layer and one output layer of processing units. Single layer perceptron only learn linear functions.

Multilayer Perceptron (Deep Neural Network): For nonlinear functions, multilayer perceptrons utilize a non-linear activation function that lets it classify data that is not linearly separable. Multilayer perceptron consists of multiple layers of hidden nodes stacked in between the layer of input & output nodes.?

Why multiple layers or hidden layers?:

As mentioned above, Artificial intelligence is inspired by our human brain where different parts of the brain are responsible for processing different aspects of information and these parts are arranged in hierarchically. As our brain receives inputs & gets processed inside the brain, each level of neurons provides insight and then information gets passed on to the next higher level. This phenomenon is captured in ANN models by hidden layers; as the number of layers increases, the learning capability of the model increases however it also slows down the computer system and hence an optimal number of layers are considered based on the computational power of the system.

The model is required to feed a tremendous amount of information/data to learn and these data are called training sets. During the training period, output of machine algorithms is compared to the human-provided output. If they are the same, the machine algorithm is validated. If it is not, it uses backpropagation (each layer’s weights are updated based on the derivative of its output w.r.t input & weight) to adjust its learning which makes a network intelligent. Once, model is trained and optimized, it’s performance is evaluated in terms of precision, recall, f1-score, mean square error, mean absolute error, R square etc. On successful evaluation, the model can be deployed for its use.

2. The Evolution of Maintenance

There are five maintenance approaches industry follows for asset management; Reactive, Preventive, Condition Based, Predictive and Prescriptive. Reactive maintenance, also known as corrective maintenance or failure based maintenance, could be an expensive approach for critical equipment as it may lead to loss of production, opportunity & ultimately customers. Preventive maintenance, also known as time-based maintenance strategy is a proactive approach which could be a reliable approach in such a situation however it might not be effective as there will still be unplanned shutdown & expensive repair that could have been avoided. This type of strategy tends to ensure safety & service maintenance by over-maintaining the asset, thus causing high economic cost. Condition-Based maintenance consists of anticipating a maintenance activity based on evidence of degradation & deviation from the normal behavior of assets[6]. Condition based monitoring is the first step towards Industry 4.0 maintenance strategy where assets/equipment/machines are continuously monitored and data is being collected either manually or through sensors while they are still running. Predictive maintenance is the advanced version of condition based approach which forecast the equipment failure in advance and alert the maintenance manager. Predictive maintenance consists of using all the information that composes and surrounds a system, and using it to predict its remaining life. Prescriptive maintenance is the most advanced knowledge based maintenance strategy which involves the integration of big data (historical & real-time data), analytics, machine learning & artificial intelligence. A prescriptive maintenance system will be a cognitive system where the system will think and perform maintenance just-in-time. The below diagram (fig-4) shows the knowledge & information that need to be collected to obtain intelligent maintenance[7].

3. Digital Twins

Digital twins is a virtual representation of a physical object, asset, processes, system or device across its life-cycle. It uses real-time data and other sources to enable learning, reasoning and dynamically recalibrating for improved decision making[6]. Digital twins integrate Internet of things, artificial intelligence, machine learning and software analytics with spatial network graphs to create living digital simulation models that update and change as their physical counterparts change. A digital twin continuously learns and updates itself from multiple sources to represent its real-time status, working condition or position.?

How does Digital Twins Work?

Digital twins were developed by NASA to operate, maintain or repair systems when they are not within physical proximity to the system. Let's think of Digital Twins as a bridge between the digital and physical world. Thousands of sensors distributed throughout the?physical assets collectively capture data along a wide array of dimensions from behavioural characteristics of equipment, work in progress such as thickness, temperature, basicity, velocity, Alkalies etc and environmental conditions within the plant. These data are continuously communicated to and aggregated by the Digital Twin system. The digital twin application continuously analyzes incoming data streams and compares it with an ideal range of acceptable data. Investigation and corrective action get triggered when the system detects trends outside the acceptable range.??

?3. The Refractory Maintenance?

Refractories are designed to withstand harsh operating temperatures and act as a protective layer for the process vessel. It plays a significant role in high temperature industrial vessels, such as, metallurgical vessels for Iron & Steel making, rotary kiln for cement industry & many other vessels for foundry, Aluminium industries. The vessels are subjected to a severe operating environment during service & their performance influenced the operating process. The campaign life of refractory depends upon thermal and thermomechanical behaviour[9].Heavy dust, temperature often higher than 1600 deg C and multiple simultaneous corrosion mechanisms limit the possibility of exact measurements and thus the predictability of refractory lifetime. In such situations, equipment operators depend heavily on the operator's experience to determine when the refractory linings are due for maintenance. This, combined with significant variability that is inherent in the process, can often lead to suboptimal performance and inconsistent results contribute to either shorter campaign life or failure of refractory.?Use of industry 4.0 not only eliminates this deficiency but also improves overall equipment/plant performance along with safety and well being of people involved in maintenance.?Equipment, process, operating parameters & maintenance practice are major drivers of refractory product selection which is predominately made by humans on the basis of knowledge and experience. This may change in the future as artificial intelligence develops further where the decision process will be increasingly assisted by self-optimizing and knowledgeable manufacturing systems. The smart factory will be equipped with a digital twin and cyber-physical system which will enable the communication between humans, machines and products alike. As they are able to acquire and process the data, they can self-control certain parts and interact with humans via interface[10].

In Iron making, blast furnace (BF) is the major equipment which requires significant refractory maintenance on a daily basis to operate it smoothly & efficiently. Although blast furnaces have been in operation since many centuries, it is still a grey area for operations & maintenance teams when it comes to inner refractories lining systems. It is nearly impossible to predict the wear & tear of inner refractory lining without shutting the furnace[11]. In such a situation, a digital twin of blast furnace can continuously monitor the Blast furnace’s refractory lining which will not only help operators to maximize the refractory campaign life but also make the process 100% safe to operate. An intelligent maintenance concept can help to improve refractory efficiency which results in higher service life, higher hot metal production and more flexibility for plant maintenance at reasonable refractory costs[10,11].

In steel making, digital twins of refractory lining of basic oxygen furnaces (BOF) can be generated to predict the refractory lifetime which will help operators to operate the BOF at optimized production level with increased safety. The laser measurement system will measure the refractory lining thickness after certain frequency (initially, it could be at every 10 heats and later every alternate heats) and send the data to Digital twins system where refractory wear model will automatically generated with the help of machine learning without any human intervention and will provide real-time wear lining pattern & balance life. This is further integrated with the gunning equipment which can automatically trigger as per the situation to repair the refractory lining.?

Conclusion

Uses of smart & intelligent technology are transforming traditional maintenance practice into smart maintenance practice AI/ML and digital twins are enabling this paradigm shift which address some of key challenges such as unforeseen downtime for relining, suboptimal and excessive maintenance practices, unsafe operations etc. The new smart technology helps to create refractory wear models, precisely identify the key wear influencing parameters, do the refractory benchmarking and automate the maintenance. This helps plant operators to predict the lifetime of lining, reduce refractory consumption and increase safety. In nutshell, AI/ML, digital twins, IIOT etc are the backbone of industry 4.0. The deployment of cutting-edge AI and machine learning technology along with digital twins will lead to massive disruption in the refractory lining maintenance[10,11].

References:

[1] Industrie 4.0 Working Group: Recommendations for implementing the strategic initiative Industrie 4.0, 2013

[2] Wikipedia: https://en.wikipedia.org/wiki/Fourth_Industrial_Revolution

[3] Forbes 2016 : Data Science Task Survey

[4] https://www.slideshare.net/arjitkantgupta/artificial-neural-networks-79614688

[5] The Digital Mine: A Four-step Approach to Predictive Maintenance 4.0

[6] IBM Blog - Digital Twins in Nutshell ?

[7] Digital Twin for maintenance: A literature review by Itxaro Errandonea, Sergio Beltran, Saioa Arrizabalaga

[8] Application of Some digital Techniques to optimize the Thermomechanical behaviour of refractory lining, Aidong Hou, Montan Universitat Leoben

[9] Medium ARTICLE - Uses of Artificial Intelligence in Refractory Maintenance, Prakash Bharati?

[10] LINKEDIN ARTICLE - Uses of Artificial Intelligence in refractory maintenance, Prakash Bharati