Bulk Microstructure Characterisation


Scientific Assurance for Future Engineering Risks (SAFER) is a group dedicated to the development of a microstructural Digital Twin for assuring the safety of high integrity components. A Digital Twin is a real-time virtual representation of a process or component, which can modify its predictions of downstream outputs by using data periodically supplied by an operator. Based on the supplied data, the Digital Twin is then able inform the operator whether the component or process downstream will be within the specifications of an application.

The aim of SAFER’s Digital Twin is to obtain safety assurances of the expected mechanical properties of a component. This is performed by supplying critical microstructural data into the Digital Twin at several key stages of the manufacturing process. The Digital Twin then incorporates the supplied data into its models and updates its predictions of the final microstructure at the end of the manufacturing process. Given the inseparable relationship between the microstructure and the mechanical properties of materials, the predicted microstructure is invariably a prediction of the mechanical properties of a component. Therefore, this Digital Twin could provide a method to assure manufacturers that their produced components are safely fulfilling the mechanical specifications of an application.

The challenges associated with the Digital Twin method, even from a microstructural perspective alone, are non-trivial. Data collected from sampled material must be representative of the overall component. Inhomogeneities within the component are almost inevitable during manufacturing, some of which can significantly change the properties of the material. Such regions must be captured to give an accurate representation of the component. However, how can this be done without sampling the entire component? Furthermore, the amount of data contained within a single microstructure is enormous. Handling and processing this data is impractical, requiring dimensionality reductions and targeted data extraction. However, the recipe for extracting this data, as well as what specific data should be harvested for the Digital Twin to function, remains elusive.

The postdoctoral role at The University of Manchester is focused on finding solutions to the challenges relating to bulk microstructural characterisation of steels and alloys. One of SAFER’s objectives involves developing a procedure for capturing the necessary microstructural data required for the Digital Twin to function. Methods for microstructural fingerprinting and categorisation are an integral part of this project, with specific focus on predicting the mechanical properties of a give set of microstructural features. To assist with the complex relationships between microstructures and mechanical properties, a new categorisation system for alloys will be developed.

Scanning electron microscopy images of heat-treated SA508 Grade III steel
Scanning electron microscopy images of heat-treated SA508 Grade III steel

Scanning electron microscopy images of heat-treated SA508 Grade III steel

Research Areas

Characterisation of microstructures is a key aspect of this research focus. Identifying the critical microstructural features that impact mechanical properties and identifying the methods by which these features can be investigated are essential for the Digital Twin. In this research focus, techniques used to analyse microstructures, and the defects within, will be compared and scrutinised. Microstructural characterisation procedures will be outlined to assist manufactures assessing the components they produce.

Metallurgical Characterisation

With the data collected from the Metallurgical Characterisation research focus, critical information must be extracted from it. This information can be tailored to the requirements of the user by using different analysis techniques and machine learning algorithms. The primary object of this research focus is to determine the best methods for extracting data to be implemented in the Digital Twin, to construct a microstructural fingerprint, and construct a bespoke microstructural categorisation methodology.


High Integrity Component Manufacturing

High Integrity Component Manufacturing
High Integrity Component Manufacturing

Manufacturing process pipeline and the associated microstructural features and methods of investigation.

Manufacturing metallic components takes place over multiple stages to achieve the correct composition, near-net shape, and microstructure. The typical stages, shown above, include melting, casting, forming, heat treating, machining, and fabrication. Each of the stages induces changes to the microstructure including dendritic formation, recrystallisation, phase transformations, precipitation, local melting and fusion, and so on. Each of these microstructural aspects needs to be carefully controlled to ensure the correct properties are achieved at the correct stage in the manufacturing.

However, the idealised microstructure for the intended application is never obtained due to the formation of defects during manufacturing. Examples of these defects can include segregation of elemental components during solidification, leading to the formation of potentially deleterious phases and inclusions. Additionally, pores, voids, and microcracks can form during most stages of processing due to dissolved gases, stress relief and so on. These defects alter the behaviour of materials by acting as stress raises to the surrounding microstructure, reducing the stress necessary to induce plastic deformation. If a critical number of defects are present, the plasticity of the material can markedly change, transitioning from ductile to brittle behaviour.

Historically, brittle fracture of metallic materials has been the cause of many major engineering disasters including the Boston Molasses Flood in 1919, the Liberty Ships of World War II, and the Silver Bridge Collapse in 1967 to name a few. The infamously unpredictable and sudden nature of brittle fracture is typically the reason why such disasters have incurred significant damage, injury, and loss. Some of these disasters spurred the inception of modern fracture mechanics, in a bid to understand how and why materials fail in different ways. Contributions from Griffith, Irwin, and others have substantially increased our awareness of fracture mechanisms and have encouraged surveillance of existing structures for potential failure using non-destructive testing and inspection. Mechanical testing, specifically toughness testing, have proven invaluable for the analysis of fracture mechanics and the identification of ductile-to-brittle transitions. However, microstructural characterisation has also been pivotal in the field of fracture mechanics, with the identification of defects, sites of crack initiation, mechanism of crack propagation, and microstructural strategies for increasing the toughness of materials.

Microstructural Characterisation Methods

EAG Laboratories Analysis Technique SMART Chart [1]
EAG Laboratories Analysis Technique SMART Chart [1]
EAG Laboratories Analysis Technique SMART Chart  []

Modern materials science enjoys a plethora of characterisation techniques for analysing different properties a material’s microstructure. The breadth of the techniques available for analysis is staggering, some of which are summarised above. Each technique produces its own output signals during an analysis to be detected and interpreted, and have their own range of resolution, detection limits and fields of view depending on the physics the technique is based on. No one technique is capable of fully characterising a microstructure. Therefore, different combinations of techniques are necessary in materials science to obtain the desired information including imaging, composition, element partitioning, crystallography, and so on.

Microstructural features can have sizes ranging from µm to nm and their contribution to the mechanical properties will vary from material to material. This provides many different recipes of characterisation techniques to investigate a material’s microstructure. However, the selected combination of techniques must be able to cover the range of scales necessary to make sound engineering judgements of the mechanical properties of a material. The selection of such characterisation techniques will indeed be microstructural-dependent. However, this provides a challenge: if the scales to be characterised are so small compared to the size of the component, how do we know the collected microstructural data is representative of the bulk microstructural character of the component? Since we cannot analyse the entirety of a macroscale component’s microstructure, due to the impracticalities and expense of doing so, we make assumptions with regards to the areas of the microstructure we analyse using multiple samples from different areas of the component.


Microstructural-Mechanical Property Relations

Consequences of fracture modes in high integrity reactor components.
Consequences of fracture modes in high integrity reactor components.

Consequences of fracture modes in high integrity reactor components.

Although the mechanical strength of a material is critical for its suitability for structural applications, a material’s toughness and response to fracture is also essential to consider during materials and engineering design. Ductile and brittle behaviour describe how a material responds to deformation. Ductile materials can absorb energy and plastically deform before fracture initiation and propagation occurs. Furthermore, fracture propagation in ductile materials is typically slow due to the material at the front of the crack tip being able to plastically deform before the crack can continue to advance. In brittle materials, however, very little energy can be absorbed by plastic deformation in the bulk material. Consequently, crack propagation occurs rapidly due to most of the mechanical energy being dissipated by forming more free surface, i.e. advancing crack growth. The sudden onset of brittle fracture is dangerous to engineering applications and is not uncommon for it to have disastrous and tragic consequences. Every effort must be made to ensure critical structures and components do not fail by brittle fracture. However, over the last century, advances in our understanding of materials and their characterisation have allowed us to study in detail the toughness and fracture toughness behaviour of engineered materials.

The toughness of a material, especially in brittle materials, can be difficult to define by a single variable since the toughness behaviour of a material can be situation dependent. Toughness, particularly in the context of fracture toughness, requires statistical and probabilistic assessments to understand how a material may behave under certain conditions. Certain features within the microstructure must be considered including inclusions, cracks, and other flaws, as well as the phases within a microstructure. The defects, like inclusions act as stress raisers to the local microstructure, which, if significant in number or size, can induce brittle fracture even in ductile materials. Furthermore, inclusions tend to de-bond from the microstructure, thereby forming microvoids which assist in crack initiation and propagation. Phases within the microstructure also need to be controlled to ensure good toughness behaviour. Coherent precipitates with blunt or rounded corners are preferred as they reduce the stress concentrations. Precipitates should be small, approximately <100µm, and should be widely dispersed to reduce stress field overlaps between precipitates. Grain sizes also influence the toughness of materials. Reducing the grain size prevents more dislocations from piling up at the boundaries, reducing the stress they exert on neighbouring grains and the chances of crack formation. Other microstructural features such as martensite or layered microstructural features can deflect the paths of cracks. This increases the distance required for a crack to induce failure, thereby increasing the toughness of the material.

Image Analysis and Categorisation

Assessing the integrity of components is an essential step of the manufacturing process. These assessments assure that the produced component is of the correct specification for the application they have been designed for. In this research focus, a guidance document, outlining procedures for investigating assumptions made by manufacturers when carrying out these assessments, will be produced. The contents of the document will also be implemented by the Digital Twin to increase its robustness in assuring the safety of high integrity components.

Representation of Microstructures in Digital Space

Digitisation of data collected by industry is at the forefront of the next industrial revolution, known as Industry 4.0. With access to increased computational power, connectivity, and automation, industries are developing new transformative methods for controlling, monitoring, and making informed judgements for their manufacturing processes. Real-time analysis of large quantities of data can inform operators whether downstream processes require modifications, whether malfunctions or production defects have occurred, or to locate where mass production inefficiencies are occurring. New computational tools have been developed over the last decade to assist Industry 4.0. These tools include machine learning methods, where algorithms can improve their performance with more task experience, neural networks, a set of algorithms inspired by neuro-physical structures in living organisms, and Digital Twins, virtual real-time representations of a physical manufacturing process.

SAFER aims to build a Digital Twin to monitor and validate the microstructure and mechanical properties of a steel component during manufacturing. The Digital Twin works by processing input microstructural data from key stages during manufacturing. The microstructural data is used to refine the Digital Twin’s predictions of the component’s downstream microstructure. With sufficient information the manufacturer can be informed whether the specifications for the component will be satisfied. For the Digital Twin to perform this task, it must be able to access a repository of microstructural data that it can use to predict the output microstructures and the associated mechanical properties. However, the microstructural data in the repository must be represented in such a way that it can be interpreted by a computer. Although pattern recognition and interpretation of, say, microstructural images by the human brain, appears trivial to us, this feat is very challenging to replicate perfectly by a computer. This research focus intends to address the challenges associated with digital storage, representation, and interpretation of microstructural data by a Digital Twin.


Microstructural Fingerprinting

Microstructural Fingerprinting
Microstructural Fingerprinting

Representing an entire microstructure in a digital format is a challenging feat. The microstructure of materials can contain large quantities of information, and if perfectly transposed into a digital format, would require huge quantities of memory space to be recorded on, let alone the interpretation of it. However, not all the information within a microstructure needs to be collected to obtain an accurate representation, or to obtain good predictions of its mechanical properties. Some of the information may be trivial with respect to the mechanical properties, whilst others can be simplified or condensed with other items of information.

Microstructural fingerprinting is process of obtaining a set of information that uniquely identifies a microstructure. The information in the fingerprint should be sufficient to reconstruct the microstructure and differentiate it from all other microstructures. However, the fingerprint should also be concise, informative, and easily transferrable in a digital format. These requirements necessitate careful selection of the most important microstructural parameters that influence the mechanical properties. Methods of data compression and efficient collation must be implemented in this process. These methods are based in computer vision disciplines such Bag of Visual Words analysis, Haralick textural features, local binary patterns, histogram orientated gradients, and so on. The precise methods and parameters to be recorded in the microstructural fingerprint are currently not defined. Part of the bulk microstructural work package in SAFER involves founding a method for constructing a microstructural fingerprint as well as establishing the critical microstructural parameters that make up the fingerprint.

Microstructural Categorisation

The end users of the Digital Twin developed in this project are primarily industrial manufacturers from small, medium, and large enterprises. The size of the enterprise typically correlates with the facilities available to perform microstructural analysis. For most end users, access to basic metallographic preparation facilities such as an optical microscope and a hardness indenter would be the minimum capabilities. Facilities such as transmission electron microscopy, and other forms of advanced characterisation are unlikely to be available for the smaller and medium-sized manufactures. Therefore, the question needs to be asked; are optical microscopes and hardness indenter facilities sufficient to construct a fingerprint and provide enough detail for the Digital Twin to produce accurate predictions?

Whilst some critical microstructural data may be obtained from basic metallographic investigatory equipment, it is unlikely that sufficient microstructural detail could be collected to facilitate a Digital Twin’s operation. This would reduce the accessibility of the Digital Twin tool to the larger, more equipped manufacturers. To mitigate this challenge, SAFER intends to develop a new categorisation method for steels and alloys. The principle of this system relies on grouping microstructures according to their similarities, for example composition, grain size, phase distributions, particle size distributions, and so on. The more similarities between two microstructures, the more similar their mechanical properties behaviour will tend to be. Organising these microstructural features as hierarchy could provide a procedure for arranging different microstructures into separate categories.

Categorising steels and alloys by this method may have some powerful advantages. If only a limited set of data can be obtained from a microstructure, for example, due the limitations of analytical equipment, the category of the investigated material can still be narrowed down with the data available. Once narrowed down, data from a closely resembling, but more thoroughly characterised, material within that category can be “borrowed” and applied to the investigated material. The data from the more thoroughly analysed microstructures provide the necessary detail for the Digital Twin to function with the investigated material. The process of data borrowing from similar microstructures allows smaller manufacturers access to the Digital Twin tools for safety validation despite not having access to more advanced characterisation instruments.

Expert Elicitation and Guidance Document

Assessment Assumptions

A critical step in manufacturing is assessing whether a produced component satisfies the standards for a particular application. This is to ensure the component will function as expected and will not undergo unexpected changes or catastrophic failure during service. These assessments include the analysis of defects within the component, making judgements whether they will hinder performance over the component’s lifetime. At every stage during manufacturing the possibility of defects forming should be assessed, as well the consequences of their presence on the downstream processes.

When undertaking the assessment procedure, assumptions must be made. The assumptions simplify the processes involved making the judgements less cumbersome and the results easier to interpret. The assumptions also make the methods more generalisable to different defect situations. However, the assessment is only as good as the assumptions it is based on. These assumptions, particularly their origin and how they are implemented, must be well understood to ensure that they are not sacrificing safety for the convenience of the assessment. This is critical knowledge that should be built into the Digital Twin architecture. With the knowledge and understanding of the assumptions better approaches to modelling processes to achieve more a representative or informed judgement could be obtained by the Digital Twin.

To assist manufacturers and the development of the Digital Twin, SAFER will build a register of these assumptions. This register will serve as the means for clarifying expert elicitations for defect assessment and experimental interrogation of the underlying assumptions during an assessment. Furthermore, when coupled with a characterisation technique register, manufacturers will be greatly informed of state-of-the-art techniques and the analyses possible. This second register is important as it ensures manufactures are kept up to date with the recent advances in well established and emerging analysis techniques, which may fulfil the analysis requirements of the manufacturer.