Methods and systems for optimising metal foil fabrication

Abstract

There is described herein a computer-implemented method for determining operating parameters of a planar flow caster in a metal foil fabrication process. The planar flow caster has a melt distribution receptacle through which a molten metal is extruded, and a cooling surface for cooling and solidifying the molten metal as the molten metal flows onto the cooling surface. The method includes capturing visual data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation, and applying at least one machine learning model to the visual data to determine one or more operating parameters of the metal foil fabrication process.

Description

Methods and Systems for Optimising Metal Foil FabricationRelated Application(s)

[0001] This application claims priority from Australian provisional application no. 2023903871, the entire contents of which are incorporates herein by reference.Technical Field

[0002] Embodiments of the present invention relate to methods and systems for monitoring and controlling a planar flow caster for metal foil fabrication.Background of Invention

[0003] Planar flow casting fabrication units (also referred to herein as planar flow casters) are employed in various industrial processes for the production of metal foils, such as shape memory alloy (SMA) metal foils and amorphous metal / alloy foils. These fabrication process involves the controlled flow of molten material onto a spinning casting wheel. The molten material subsequently solidifies to form the desired metal foil product. Efficient and precise control of operating parameters, such as temperature and speed of the casting wheel are important for the achievement of desired material properties and product quality of the metal foil.

[0004] However, several challenges exist in the operation of conventional planar flow casting fabrication units which could impede their operational effectiveness. One of the issues relates to the difficulty of accurately knowing the operating parameters associated with the planar flow caster. The extremely high temperatures present in the planar flow caster can often pose an obstacle to the placement of sensors and measurement instrumentation required for precise monitoring and control. For example, the temperature of the molten metal being extruded by the planar flow caster may be more than 600°C. These elevated temperatures not only limit the lifespan of sensors but also induce wear and degradation of their components, compromising their reliability and accuracy. In addition, sensors and instruments designed to operate in such elevated temperature environments are also extremely costly.

[0005] Furthermore, the placement of sensors and measurement instrumentation at required locations in the planar flow caster may be intrusive, which may undesirably affect the operation of the planar flow caster. Moreover, precision sensors that are capable of operating in high temperatures of a planar flow caster may be expensive.

[0006] Conventionally, the determination of control parameters for the planar flow caster and other variables associated with the fabrication process, such as properties of the metal material used for the foil product, design of a nozzle for directing the molten material onto the spinning casting wheel requires manual processes of trial and error. This may be costly and time consuming.

[0007] Embodiments of the invention may provide methods and systems for monitoring and controlling a planar flow caster for metal foil fabrication which overcomes or ameliorates one or more of the disadvantages or problems described above, or which at least provides the consumer with a useful choice.

[0008] A reference herein to a patent document or any other matter identified as prior art, is not to be taken as an admission that the document or other matter was known or that the information it contains was part of the common general knowledge as at the priority date of any of the claims.Summary of Invention

[0009] According to one aspect of the invention, there is provided a computer- implemented method for determining operating parameters of a planar flow caster in a metal foil fabrication process, the planar flow caster having a melt distribution receptacle through which a molten metal is extruded, and a cooling surface for cooling and solidifying the molten metal as the molten metal flows onto the cooling surface, the method including capturing visual data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation, applying at least one machine learning model to the visual data to determine one or more operating parameters of the metal foil fabrication process.

[0010] There is also disclosed herein, a computer-implemented method for determining operating parameters of a planar flow caster in a metal foil fabrication process, the planar flow caster having a melt distribution receptacle through which a molten metal is extruded, and a cooling surface for cooling and solidifying the molten metal as the molten metal flows onto the cooling surface, the method including capturing visual data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation, applying at least one machine learning model to the visual data to determine one or more operating parameters of the metal foil fabrication process, wherein the one or more operating parameters includes any one or more of a thermal distribution map associated with the SMA foil fabrication process, a variable distance between the melt distribution receptacle and the cooling surface, and a flow velocity of the molten metal from the melt distribution receptacle towards the cooling surface.

[0011] Embodiments of the present invention therefore uses visual data to effectively monitor important operating parameters of the planar flow caster without the need for costly sensors or other measurement instrumentation. Indeed, the temperature inside a chamber of the planar flow caster may exceed 1500°C. For example, the temperature inside the chamber may reach 2500°C or more. It is often not practical to use conventional temperature sensors and measurement instrumentation in such extreme heating environments. By relying on visual data analysis and advanced computer vision techniques, embodiments of the present invention may provide cost effective, robust and reliable monitoring methods and systems which can be used in harsh operating environments, where conventional sensor and instrument based data acquisition techniques cannot be deployed.

[0012] The method may further include pre-processing the video data including any one or more of image segmentation, edge detection or boundary detection, feature extraction and object recognition.

[0013] Typically, the video data may be pre-processed into a sequence of image frames. The image frames may be in colour (RGB) or greyscale. Each image frame may be time stamped, and further processed to determine the relevant operating parameters at a corresponding point in time. The sequence of image frames may be processed iteratively so as to determine the one or more operating parameters of the planar flow caster in time series.

[0014] The processing of the video data to determine the one or more operating parameters may be carried out during or post operation of the planar flow caster. In one embodiment, the one or more operating parameters of the SMA foil fabrication process may be determined in real-time or near real-time based on the video data.

[0015] The one or more operating parameters may include any one or more of a thermal distribution map associated with the metal foil fabrication process, a variable distance between the melt distribution receptacle and the cooling surface, a flow velocity or flow rate of the molten metal from the melt distribution receptacle towards the cooling surface, and speed of a casting wheel providing the cooling surface.

[0016] The thermal distribution map may provide thermal distribution information for a chamber of the planar flow caster. In some embodiments, the thermal distribution map may provide thermal distribution information for a region of interest within the chamber of the planar flow caster. For example, the thermal distribution map may include a thermal distribution of the melt distribution receptacle, molten metal and cooling surface during operation of the planar flow caster.

[0017] The visual data may include image data, video data or a combination thereof. In one embodiment, the visual data may include a sequence of images captured at predetermined time intervals during operation of the planar flow caster. In another embodiment, the visual data may include live video data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation. As mentioned, the video data may be pre-processed into a sequence of image frames in time series.

[0018] At least one of the operating parameters may be a thermal distribution map. In this embodiment, the method may further include pre-processing the visual data includingclustering pixels based on mean cluster centroids to generate pixel clusters. Moreover, the step of determining may include determining one or more temperature values corresponding to each pixel cluster using a supervised machine learning model. The method may further include generating the thermal distribution map based on the temperature values and pixel clusters.

[0019] In one embodiment, the method may include determining an operating temperature range corresponding to each image frame generated from the visual data, determining a plurality of temperature zones based on the operating temperature range, generating pixel clusters and assigning each pixel cluster to a corresponding temperature zone. The thermal distribution map may be generated based on the pixel clusters and corresponding temperature zones.

[0020] Typically, a temperature sensor may be provided for determining a maximum temperature threshold for the operating temperature range.

[0021] Any suitable number of temperature zones may be used. In one embodiment, the number of temperature zones may be predetermined based on a desired resolution for the thermal distribution map. For example, 10 temperature zones may be used to generate the thermal distribution maps. In some embodiments, more than 10 temperature zones may be used if a higher resolution thermal distribution map is required. Alternatively, less than 10 temperature zones may be used if a lower resolution thermal distribution map would be sufficient.

[0022] At least one of the operating parameters may be a variable distance between the melt distribution receptacle and the cooling surface. In this embodiment, the method may further comprise pre-processing the visual data including determining a first boundary corresponding to the melt distribution receptacle, and determining a second boundary corresponding to the cooling surface.

[0023] Typically, the cooling surface is provided on a casting wheel of the planar flow caster. During operation, the casting wheel rotates as the molten metal flows onto the cooling surface. The molten metal solidifies on the cooling surface to form the final metal foilproduct. As such, the first boundary may correspond to a lower boundary of the melt distribution receptacle, and the second boundary may correspond to the plane of the cooling surface (or an upper boundary of the casting wheel).

[0024] In practice, limited illumination inside the chamber of the planar flow caster may create conditions of low visibility, rendering the utilisation of image data for delineating object boundaries challenging. Indeed, the reduced lighting conditions within the chamber of the planar flow caster can often present obstacles to the accurate determination of object boundaries through image analysis. In embodiments of the present invention, it was discovered that a reflection of the melt distribution receptacle on the cooling surface of the casting wheel can be used to accurately identify the relevant upper boundary of the casting wheel for the purposes of calculating the variable distance between the melt distribution receptacle and the cooling surface.

[0025] Accordingly, determining the second boundary may include determining the second boundary based on a reflection of melt distribution receptacle on the cooling surface.

[0026] In some embodiments, computer vision techniques for object detection and recognition, edge detection and contour detection may be used to locate the melt distribution receptacle and its reflection to determine the first and second boundaries.

[0027] The first boundary and the second boundary may be determined using one or more unsupervised machine learning models.

[0028] The variable distance between the melt distribution receptacle and the cooling surface may be determined based on a distance between the first boundary and the second boundary.

[0029] The distance between the first boundary and the second boundary may be determined using a supervised machine learning model. The supervised machine learning model may be trained based on the spatial distance in pixels from image data against physical distance between the melt distribution receptacle and cooling surface as measured during an experimental setup. Measurements of physical distances associated with components of the planar flow caster may be used to calibrate the corresponding spatial distances obtained from the image data.

[0030] At least one of the operating parameters may be a flow velocity of the molten metal from the melt distribution receptacle towards the cooling surface. In this embodiment, the method may further comprise pre-processing the visual data, which may include determining a region of interest between the melt distribution receptacle and the cooling surface, wherein a boundary of the cooling surface may be determined based on a reflection of the melt distribution receptacle on the cooling surface.

[0031] In some embodiments, the method of determining flow velocity of the molten metal and the method of determining the variable distance are interlinked. For example, once the first and second boundaries are determined, the first and second boundaries can be used to determine a portion of each image frame corresponding to the molten metal. Pixel based measurements associated with the portion of each image frame corresponding to the molten metal may be used to determine flow velocity.

[0032] In particular, the planar flow caster may be pre-configured such that molten metal will only flow from the melt distribution receptacle after the variable distance between the melt distribution receptacle and the cooling surface reaches a predetermined distance (e.g. 2mm). As such, once the variable distance between the first and second boundaries reaches the predetermined distance, it may be determined that dark pixels between the first and second boundaries correspond to the molten metal.

[0033] In some embodiments, determining the flow velocity of the molten metal may include determining an intensity or colour threshold for image pixels representative of the molten metal, estimating a width of the pixels at or above the threshold within the region of interest, and estimating a height of the pixels at or above the threshold within the region of interest.

[0034] Optionally, determining the flow velocity of the molten metal may include calculating the flow velocity based on the estimated width and height.

[0035] At least one of the operating parameters may be a flow rate of the molten metal from the melt distribution receptacle towards the cooling surface. The method may further comprise comparing frames of visual data and determine changes in pixel intensity values, identifying regions of interest indicative of flow of molten metal, and determining the flow rate based on area measurements of the identified regions of interest.

[0036] At least one of the operating parameters may be angular velocity of the casting wheel providing the cooling surface. The method may further comprise determine circular features of the casting wheel based on edge detection, generate velocity vectors based on the circular features, determine angular velocity of the casting wheel based on the velocity vectors.

[0037] Optionally, the computer-implemented method may be configured to determine one or more quality parameters associate with the metal foil produced by the planar caster. The one or more quality parameters may include one or more dimensions of the metal foil. The one or more dimensions may include either one or both of a width and thickness.

[0038] Optionally, the metal foil may be a shape memory alloy (SMA). Alternatively, the metal foil may be an amorphous alloy foil.

[0039] According to another aspect of the invention, there is provided a method of determining control parameters for controlling operations of a planar flow caster in a metal foil fabrication process, wherein the planar flow caster includes a heating coil for heating a mass of metal during a metal melting process, and a casting wheel, the casting wheel providing the cooling surface, the method including determining one or more operating parameters according to the method for determining operating parameters of a planar flow caster in a metal foil fabrication process as described herein, and determining one or more control parameters based on the one or more determined operating parameters to control operations of the planar flow caster, wherein the control parameters includes any one or more ofa temperature of the heating coil, a rotating speed of the casting wheel, a descending speed of the melt distribution receptacle, and a pressure generated during the metal melting process.

[0040] Typically, control parameters such as the temperature of the heating coil, rotating speed of the casting wheel, descending speed of the melt distribution receptacle may be directly controllable via a user interface or control panel of the planar flow caster. Other control parameters such as the pressure generated during the metal melting process may be varied by altering related parameters such as the mass of the metal material placed into the planar flow caster.

[0041] There is also disclosed herein, a method of operating a planar flow caster in a metal foil fabrication process, the method including a method for determining one or more operating parameters as described herein.

[0042] There is also disclosed herein, a method of operating a planar flow caster in a shape memory alloy (SMA) foil fabrication process, the method including a method for determining one or more control parameters as described herein.

[0043] There is also disclosed herein, a metal foil fabricated in accordance with a method of operating a planar flow caster in a metal foil fabrication process as described herein.

[0044] The metal foil may be a shape memory alloy (SMA) foil. A width of the foil may be about 30mm to 100mm. In particular, the width of the foil may be more than 30mm, 40mm, 50mm, 60mm, 70mm, 80mm, or 90mm.

[0045] According to yet another aspect of the invention, there is provided a planar flow caster, including a heating coil for heating a mass of metal during a metal melting process to create a molten metal, a melt distribution receptacle through which the molten metal is extruded, anda casting wheel, the casting wheel having a cooling surface for cooling and solidifying the molten metal as it flows onto the cooling surface from the melt distribution receptacle, and a camera operatively configured to capture visual data of the melt distribution receptacle, molten metal and cooling surface during operation.

[0046] The planar flow caster may further include a process monitoring module operatively configured to process the visual data via any one or more of image segmentation, edge detection or boundary detection, feature extraction and object recognition, and apply at least one machine learning models to the visual data to determine one or more operating parameters of the planar flow caster in real-time or near real-time, the one or more operating parameters including any one or more of a thermal distribution map of the metal foil fabrication process including a thermal distribution of the melt distribution receptacle, molten metal and cooling surface during operation of the planar flow caster, a variable distance between the melt distribution receptacle and the cooling surface, and a flow velocity of the molten metal from the melt distribution receptacle towards the cooling surface.

[0047] The visual data may include live video data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation.

[0048] In one embodiment, the process monitoring module may be operatively configured to generate a thermal distribution map based on the visual data. In particular, the process monitoring module is operatively configured to generate the thermal distribution map by clustering pixels in the visual data based on mean cluster centroids to generate pixel clusters, and determining one or more temperature values corresponding to each pixel cluster using a supervised machine learning model, andgenerating the thermal distribution map based on the temperature values and pixel clusters.

[0049] In one embodiment, the process monitoring module may be operatively configured to determine a variable distance between the melt distribution receptacle and the cooling surface.

[0050] The process monitoring module may be operatively configured to determine a variable distance between the melt distribution receptacle and the cooling surface by processing the visual data to determine a first boundary corresponding to the melt distribution receptacle, and a second boundary corresponding to the cooling surface.

[0051] The second boundary may be determined based on a reflection of melt distribution receptacle on the cooling surface.

[0052] The process monitoring module may determine the first boundary and the second boundary by using one or more unsupervised machine learning models.

[0053] The process monitoring module may determine the variable distance between the melt distribution receptacle and the cooling surface by determining a distance between the first boundary and the second boundary.

[0054] The process monitoring module may determine the distance between the first boundary and the second boundary by using a supervised machine learning model.

[0055] The process monitoring module may be operatively configured to determine a flow velocity of the molten metal from the melt distribution receptacle towards the cooling surface.

[0056] The process monitoring module may determine the flow velocity by processing the visual data to determine a region of interest between the melt distribution receptacle and the cooling surface. A boundary of the cooling surface may be determined based on a reflection of the melt distribution receptacle on the cooling surface.

[0057] In one embodiment, the process monitoring module may determine the flow velocity by determining an intensity or colour threshold for image pixels representative of the molten metal, estimating a width of the pixels at or above the threshold within the region of interest, and estimating a height of the pixels at or above the threshold within the region of interest.

[0058] The process monitoring module may determine the flow velocity by calculating the flow velocity based on the estimated width and height.

[0059] The planar flow caster may further include a closed-loop control module. The closed-loop control module may be operatively configured to determine one or more control parameters to control operations of the planar flow caster based on one or more operating parameters determined by the process monitoring module, wherein the control parameters include any one or more of a temperature of the heating coil, a rotating speed of the casting wheel, a descending speed of the melt distribution receptacle, and a pressure generated during the metal melting process.

[0060] According to a further aspect of the invention, there is provided a metal foil fabricated using the planar flow caster as described herein.

[0061] In one embodiment, the metal foil may be an SMA foil and a width of the foil may be between about 30mm to 100mm. In particular, the width of the foil may be more than 30mm, 40mm, 50mm, 60mm, 70mm, 80mm, or 90mm.

[0062] There is also disclosed herein, a process monitoring and control system for a planar flow caster, the system including a process monitoring module being operatively configured to carry out a method for determining operating parameters of a planar flow caster in a metal foil fabrication process as described herein.

[0063] The process monitoring and control system may further include a closed-loop control module, the closed-loop control module being operatively configured to determine one or more control parameters to control operations of the planar flow caster based output from the process monitoring module, wherein the control parameters include any one or more of a temperature of the heating coil, a rotating speed of the casting wheel, a descending speed of the melt distribution receptacle, and a pressure generated during the metal melting process.

[0064] In order that the invention may be more readily understood and put into practice, one or more preferred embodiments thereof will now be described, by way of example only, with reference to the accompanying drawings.

[0065] It will be appreciated that for simplicity and clarity of illustration, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or analogous elements.Brief Description of Drawings

[0066] Figure 1A is a flow diagram illustrating a method of determining operating parameters and control parameters in accordance with an embodiment of the invention.

[0067] FIGURE IB is a schematic diagram illustrating an overall system including a planar foil caster, a processing monitoring module and a control module according to an embodiment of the invention.

[0068] FIGURE 2A is a flow diagram illustrating a method of generating a thermal distribution map according to an embodiment of the invention.

[0069] FIGURE 2B is a schematic diagram further illustrating the method of generating a thermal distribution map as shown in Figure 2A.

[0070] FIGURES 3A to 3C are schematic diagrams illustrating the interaction between the heating coil, melt distribution receptacle and casting wheel of a planar flow caster according to an embodiment of the invention.

[0071] FIGURE 3D is a flow diagram illustrating a method of generating a variable distance between the melt distribution receptacle and the casting wheel according to an embodiment of the invention.

[0072] FIGURE 4A is a flow diagram illustrating a method determining a flow velocity of the molten metal according to an embodiment of the invention.

[0073] FIGURES 4B is a graph illustrating example time series results for the variable distance recorded during an experiment executing the method shown in Figure 3D.

[0074] FIGURES 4C is a graph illustrating example time series results for the flow velocity recorded during an experiment executing the method shown in Figure 4A.

[0075] FIGURE 5A is a schematic diagram of a planar flow caster suitable for fabricating amorphous metal foils according to an embodiment of the invention.

[0076] FIGURE 5B is a flow diagram illustrating a system suitable for fabricating amorphous metal foils according to an embodiment of the invention.

[0077] FIGURE 6A is a flow diagram illustrating a method of determining a flow rate of the molten metal according to an embodiment of the invention.

[0078] FIGURES 6B to 6G are exemplary image frames (pre- and post-processing) captured by a camera during fabrication of an amorphous metal foil according to an embodiment of the invention.

[0079] FIGURE 7 is a flow diagram illustrating a method of determining a speed of the spinning wheel according to an embodiment of the invention.

[0080] FIGURE 8A is a flow diagram illustrating a method of determining dimensions of the foil during the PFC process according to an embodiment of the invention.

[0081] FIGURES 8B to 8C are exemplary image frames captured by a camera during fabrication of a metal foil according to an embodiment of the invention.

[0082] FIGURES 8D and 8E are exemplary image frames captured by a camera during fabrication of a metal foil which are overlaid with gradient maps of the percentage thickness variation of the metal foil determined in accordance with an embodiment of the invention.Detailed DescriptionMethod Overview

[0083] Figure 1A is a flow diagram which broadly illustrates a method 10 for determining operating parameters of a planar flow caster in a metal foil fabrication process.

[0084] At step 12, visual data of the planar flow caster may be provided to a monitoring module associated with the planar flow caster. The monitoring module may include one or more trained models such as trained statistical and / or machine learning models for receiving the input visual data and predicting one or more operating parameters of the planar flow caster. Alternatively or in combination, one or more operating parameters of the planar flow caster may be determined by the monitoring module based on vision processing techniques. The visual data may be in the form of video data and / or image data. Typically, the visual data may be captured by a camera. The camera may be positioned to capture visual data of at least a portion of the planar flow casting (PFC) process. The planar flow casting (PFC) process involves rapidly cooling a molten alloy to form a thin, continuous ribbon of metal foil. In this process, molten alloy is extruded from a melt distribution receptacle onto a fast-spinning, casting wheel providing a cooling surface. As the molten alloy contacts the cooling surface of the casting wheel, it rapidly solidifies at a high cooling rate.

[0085] For shape memory alloy (SMA) foils, such as nickel-titanium-based alloys, this rapid cooling preserves the desired microstructural features that enable the alloy's shape memory and super elastic properties. Typically, the rapid quenching prevents crystallisation, allowing the formation of a fine-grained or partially amorphous microstructure, which can later be heat-treated to optimize phase transformation characteristics.

[0086] For amorphous alloy foils, also commonly referred to as metallic glasses, the PFC process enables the achievement of an amorphous structure in the metal foil. By cooling the molten metal alloy faster than its crystallisation rate, the atomic structure remains disordered, providing the material with unique properties, such as high strength and corrosion resistance, due to the lack of crystal boundaries. Further discussion of the PFC process for a shape memory alloy (SMA) foil and an amorphous metal foil will be described in further detail below with reference to Figures IB and 5.

[0087] The camera may be positioned to capture visual data of the movement of molten metal from the melt distribution receptacle to the cooling surface. Moreover, the camera may capture visual data comprising at least a portion of the melt distribution receptacle, the flow of molten metal from the melt distribution receptacle onto the cooling surface, and at least a portion of the cooling surface and casting wheel.

[0088] As mentioned, one or more statistical and / or machine learning models may be used to receive the visual data captured by the camera in step 12. Moreover, any suitable statistical and / or machine learning models may be used. The statistical and / or machine learning models may be trained models. For example, recurrent neural networks (RNN) and / or convolutional neural networks (CNN) may be used. Typically, the statistical / machine learning models have been trained using experimental data during experimental PFC processes as discussed in further detail below.

[0089] At step 14, the one or more statistical and / or machine learning models determine one or more operating parameters based on the input visual data. As used herein, the term 'operating parameters' refer to operating parameters of a planar flow caster in a metal foil fabrication process, which typically include variables that govern the operation of the planar flow casting process. The operating parameters may include any one or more of a thermal distribution map associated with the metal foil fabrication process, a variable distance between the melt distribution receptacle and the cooling surface, a flow velocity or flow rate of the molten metal from the melt distribution receptacle towards the cooling surface, and speed of a casting wheel providing the cooling surface. Optionally, the trained model may also determine one or more quality parameters based on the input visual data at step 14. As used herein, the term "quality parameter" refer to parameters that provide an indication ofthe quality of the metal foil as it is solidified on the casting wheel. The quality parameters may include dimensions of the metal foil and the percentage variation in thickness across the width of the metal foil, as discussed in further detail herein with reference to Figures 8A to 8E.

[0090] Optionally, at step 16, control parameters are determined by a control module associated with the planar flow caster, based on the one or more operating parameters from step 14. The control module and the monitoring module may be provided as separate modules, or a single integrated module. As used herein, the term 'control parameters' refer to variables that may be used to control operations of a planar flow caster in a metal foil fabrication process. In other words, the control parameters may refer to adjustable settings for managing and regulating the planar flow casting process. Some example control parameters may include a temperature of the heating coil, a rotating speed of the casting wheel, a descending speed of the melt distribution receptacle, and a pressure generated during the metal melting process. In some embodiments, a control parameter may include a temperature of the cooling surface of the casting wheel. As described in further detail below, the one or more operating parameters obtained from step 14 may be used to determine or adjust one or more control parameters of the planar flow caster to improve the PFC process, which may provide continuous quality monitoring and improvement for the PFC process and achieve higher quality metal foil products.

[0091] Optionally, at step 18, the control module may adjust one or more control parameters of the planar flow caster based on the determined control parameters in step 16. For example, the control module may compare the determined control parameters in step 16 against actual control parameters used to operate the planar flow caster and adjust the one or more actual control parameters if the difference exceeds a predetermined threshold. As such, in some instances, the control module may operate as a feedback control module for the planar flow caster.

[0092] Embodiments of the invention will now be described in relation to the fabrication of shape memory alloy (SMA) foil and amorphous metal foil. It will be understood that the methods and systems described herein may also be applicable to the fabrication of other types of metal foils.Shape Memory Alloy (SMA) Foil Fabrication

[0093] A system 100 for improved shape memory alloy (SMA) foil fabrication according to an embodiment of the invention is illustrated in Figure IB. The system 100 includes a planar flow caster 102 and a process monitoring module 104 for determining operating parameters of the planar flow caster 102. Optionally, the system 100 may further include a control module 106 (such as a closed-loop control module) operatively configured to determine one or more control parameters to control operations of the planar flow caster 102 based output from the process monitoring module 104.

[0094] In some embodiments, the planar flow caster 102, process monitoring module 104 and control module 106 may be provided as separate discrete components of the system 100. In some embodiments, the planar flow caster may include an integrated process monitoring module 104. Moreover, the monitoring module 104 and control module 106 may be integrated into a single module, or provided separately.Planar Flow Caster

[0095] In the embodiment illustrated in Figure IB, the planar flow caster 102 includes a chamber 110 (often referred to as an atmosphere chamber). The chamber 110 contains a heating coil 108 for heating a mass of SMA during an SMA melting process to create a molten metal (not shown). The chamber 110 further contains a melt distribution receptacle 112 through which the molten metal is extruded, and a casting wheel 114. The casting wheel 114 spins during operating and has a cooling surface 116 for cooling and solidifying the molten metal. As the molten metal flows onto the cooling surface 116 of the spinning casting wheel 114 from the melt distribution receptacle 112, it cools and forms a thin foil (herein referred to as the SMA foil). Typically, the melt distribution receptacle 112 includes a crucible (e.g. a carbon crucible) configured to hold the molten metal, and a nozzle generally located at a base of the crucible to direct the molten metal onto a cooling surface 116 of the casting wheel 114. The planar flow caster 102 further includes a camera 118 operatively configured to capture visual data of the melt distribution receptacle 112, molten metal and cooling surface 116 during operation.

[0096] Typically, the chamber 110 retains heat provided by the heating coil 108 so as to provide a high temperature environment to produce the molten metal. In some embodiments, the temperature of the molten metal is heated to 600 degrees centigrade and above. The planar flow caster 102 may further include a temperature sensor 120 for measuring the temperature within the chamber 110. The temperature sensor 120 may measure a localised temperature within the chamber 110. Sensor data from the temperature sensor 102 may be transmitted to the process monitoring module 104 and / or control module 106 to facilitate determining one or more operation parameters, or control parameters as described in further detail below.

[0097] As it will be appreciated, conventional temperature sensors may be ill-suited for use in such high temperature environments due to their inherent limitations. For example, the materials used in the sensors may not be able to withstand such extreme temperatures. Operating such sensors outside the appropriate temperate thresholds may lead to inaccurate measurements, degradation, malfunctioning or complete failure. On the other hand, specialised temperature sensors for high-temperature applications may be relatively expensive. Moreover, temperature sensors are typically only suitable for providing a single temperature reading for each localised area that a temperature sensor is mounted. As such, high resolution thermal distribution information may be difficult to obtain based on sensor readings alone. As described in further detail below, the visual data captured by the camera 118 can be used to provide thermal distribution information associated with the planar flow caster 102 and chamber 110, which overcomes some of these limitations with conventional temperature sensors.

[0098] As illustrated in Figure IB, the camera may be mounted to the chamber 110. The chamber 110 may define an opening 122 generally aligned with a lens of the camera 118 so that the camera 118 can capture visual data from inside the chamber 110.

[0099] The visual data captured by the camera 118 may include one or both of image data and video data. In one embodiment, the visual data includes video data which captures movement of the melt distribution receptacle 112 towards the cooling surface 116 and the flow of molten metal from the melt distribution receptacle 112 on to the cooling surface 116.In particular, the video data may be live video data to enable real-time or near real-time monitoring of the planar flow caster 102 operating parameters.

[0100] As described in further detail below, the visual data captured by the camera 118 can be used by the process monitoring module 104 to determine any one or more of a plurality of operating parameters of the planar flow caster 102.

[0101] Advantageously, the system 100 employs visual data and computer vision to monitor a wide range of different operating parameters of the planar flow caster 102, without the need to use conventional sensors, such as conventional temperature sensors and the like.Process Monitoring Module

[0102] As illustrated in Figure IB, the process monitoring module 104 is operatively configured to determine one or more operating parameters of the planar flow caster 102. The operating parameters include a thermal distribution map 124 associated with the SMA foil fabrication process, a variable distance 126 between the melt distribution receptacle 112 and the cooling surface 116, and a flow velocity 128 of the molten metal from the melt distribution receptacle 112 towards the cooling surface 116.

[0103] Typically, the visual data captured by the camera 118 is pre-processed 130 and calibrated 132 by the process monitoring module 104. This pre-processing may include for example, image segmentation, edge detection or boundary detection, feature extraction and object recognition.

[0104] In particular, the video data may be pre-processed and calibrated into a sequence of image frames 350. The image frames 350 may be in colour (RGB) or greyscale. Each image frame 350 may be time stamped, and further processed to determine the relevant operating parameters 124, 126, 128 at the corresponding point in time. The sequence of image frames may be processed iteratively so as to determine the one or more operating parameters 124, 126, 128 of the planar flow caster in time series.

[0105] The pre-processed and calibrated visual data is used to determine the operating parameters 124, 126, 128 using one or more machine learning and / or statistical models 200, 300, 400.

[0106] Typically, the one or more operating parameters 124, 126, 128 of the SMA foil fabrication process are determined in real-time or near real-time based on the video data, for example live video data. The live video data may be video data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation. Capturing real-time or near real-time video data may enable the determination of operating parameters 124, 126, 128 in real-time or near real-time such that appropriate control parameters may be determined in a timely manner.

[0107] In the method 200 of generating a thermal distribution map 202, 204, 206, each pre-processed image frame may be further processed to generate clusters of pixels corresponding to a plurality of temperature zones. A total number of temperature zones may be predetermined based on a desired resolution of the output thermal distribution map 202, 204, 206. For example, a higher number of temperature zones may provide a higher resolution thermal distribution map 202, 204, 206 and vice versa. In one example, the predetermined number of temperature zones may be 10.

[0108] A temperature sensor reading from sensor 120 corresponding to each image frame may be used to set a maximum temperature (also referred to herein as a dynamic thermal threshold) for the temperature zones. The maximum temperature together with the pixel clusters may be used as input to a pre-trained machine learning model to determine the appropriate temperature values corresponding to each cluster of pixels. The clusters of pixels and their corresponding temperature values may be used to generate the thermal distribution maps 202, 204, 206. Further detail of the method 200 will be described below with reference to Figures 2A and 2B.

[0109] In the method 300 of determining a variable distance 126 between the melt distribution receptacle 112 and the cooling surface 116, one or more image processing or computer vision techniques may be used to locate pixels defining the melt distribution receptacle 112 and its reflection 134. The reflection 134 may be used to locate the cooling surface 116 in each image frame 350.

[0110] Computer vision techniques may be used to delineate the pixels corresponding to the melt distribution receptacle 112 to determine a first boundary, and delineate the pixels corresponding to the reflection 134 so as to determine a second boundary. The first boundary may be at a lower extremity of the pixels representing the melt distribution receptacle 112, and the second boundary may be at an upper extremity of the pixels representing the reflection 134. A separation between the first and second boundaries may be used to determine the variable distance operating parameter 126. Further detail of method 300 will be described below with reference to Figures 3A to 3D.

[0111] The method 400 of determining a flow velocity 128 of the molten metal may be interlinked with the method 300 of determining the variable distance 126. In method 400, once a distance between the first and second boundaries reach a predetermined value (e.g. corresponding to a physical distance of 2mm), dark pixels between the first and second boundaries may be considered as corresponding to the molten metal. Pixel based measurements associated with the portion of each image frame corresponding to the molten metal may be used to determine flow velocity. Further detail of method 400 will be described below with reference to Figures 4A to 4C.Thermal Distribution Map

[0112] A method 200 of generating a thermal distribution map will now be described with reference to Figures 2A and 2B. Example thermal distribution maps 202, 204, 206 generated by the method 200 are illustrated. The thermal distribution map may include a thermal distribution 206 of the chamber 110, or any part of the chamber 110. In some embodiments, the thermal distribution map 202, 204 may provide heat distribution of a focused area of interest 228. The focused area of interest 228 may include a thermal distribution of the melt distribution receptacle 112, molten metal and cooling surface 116 during operation of the planar flow caster 102.

[0113] At step 208, the video data is pre-processed into a series of RGB coloured image frames.

[0114] At step 210, the coloured RGB image frames are converted to grayscale image frames.

[0115] At step 212, temperature measurements (for example, obtained from sensor 120) within the chamber 110 can be used to provide a maximum temperature value for the chamber 110 or a focused area of interest within the chamber 110 for generation of the thermal distribution map 202, 204, 206. In one example, a predetermined number of temperature zones (e.g. 10 temperature zones) spread across a specific temperature range are provided. For instance, the temperature range may be Tminto Tmax, wherein Tminmay be a minimum temperature of the chamber 110, and Tmaxis the temperature measurement from sensor 120 (also referred to herein as a dynamic thermal threshold). In one embodiment, Tminmay be 300°C. The temperature zones may be assigned to corresponding intensity levels (ranging from values 0-255) for the image frames such that Tmincorresponds to a minimum intensity level, lmin (e.g. lmin = 150) and Tmaxcorresponds to maximum intensity level, lmax(e.g. Imax= 255). Each image frame may have a different value for Tmin, Tmax, lmin, lmax, which results in different temperature zones and corresponding pixel values. One or more machine learning models may be used to dynamically classify each pixel of an image frame and assign it to a corresponding temperature zone. At step 214, the maximum temperature value determined in step 212 is used to provide a dynamic thermal threshold as described above. The dynamic thermal threshold can be used to determine upper and lower limits of the different temperature zones to which different corresponding pixels in the image frame can be assigned.

[0116] At step 216, the process monitoring module 104 identifies groups of adjacent or neighbouring pixels (e.g. groups of greater than 8 pixels) that have higher brightness or intensity values compared to the surrounding pixels.

[0117] At step 218, pixel clusters are generated based on mean cluster centroids. In particular, clustering method is applied to form pixel clusters corresponding to the different temperature zones based on respective brightness values of the pixels (0-255).

[0118] At step 220, the pixel clusters generated in step 218 and the maximum temperature value from step 212 are provided as input to a pre-trained machine learning model for determining thermal zones. The machine learning model for determining thermal zones and training process will be described in further detail below.

[0119] At step 222, the machine learning model provides as output predicted thermal zones and maximum temperature values for each of the temperature zones.

[0120] At step 224, the thermal zones in step 222 are applied to all pixels of each greyscale image frame to generate a thermal distribution map.

[0121] At step 226, the clustering of pixels is iteratively refined to improve the machine learning model in Figure 220.

[0122] During operation, the temperature inside the chamber 110 is dynamically changing. As such, the upper and lower limits of the temperature zones and the upper and lower limits of intensity values of the pixels in the image frames are also constantly changing. Based on the sensor data from temperature sensor 120 in time series, the machine learning model may be provided with a new set of temperature zones as training targets and correlated pixel intensity ranges as training inputs for further training. In particular, the machine learning model may be iteratively re-trained to refine the clustering models, to produce a new set classified groups or clusters of pixels, which are labelled with a pointer to a corresponding temperature zone. Once a group of pixels are classified accordingly, each cluster of pixels is labelled with an individual temperature value to generate a thermal distribution map corresponding to each input image frame.

[0123] In some embodiments one or more unsupervised computer vision techniques and unsupervised pixel clustering methods (e.g. K-means clustering) may be used to generate the pixel clusters and corresponding dynamic temperature zones.

[0124] Generally, the process of establishing and configuring the machine learning model to determine thermal distribution may include a series of steps including data collection, data pre-processing, model training, model validation, prediction and visualisation, and finally optimisation and decision making.

[0125] During data collection, a comprehensive dataset of temperature measurements is compiled. This temperature data can be collected using a plurality of temperature sensors (e.g. temperature sensor 110 and other like sensors) strategically placed within the chamber 110, capturing temperature readings at various locations and time intervals. Typically, thetemperature dataset covers a wide range of operating conditions and casting scenarios to ensure the model's accuracy and generalisability.

[0126] During data pre-processing, the collected temperature dataset is pre-processed to remove noise, outliers, and inconsistencies. Data pre-processing techniques, such as data cleaning, normalization, and feature engineering, may be applied to enhance the quality and usefulness of the temperature dataset.

[0127] During model training, a machine learning model, such as a deep learning neural network, can be trained using the pre-processed temperature dataset. The machine learning models can be trained to find patterns in data and learn the complex relationships between input variables (e.g. image pixels) and the corresponding temperature distributions. Typically, the training process involves adjusting the model's parameters to minimize the difference between the predicted and actual temperature values. In some embodiments, the machine learning models may be trained using image pixels and corresponding temperature datasets obtained under different casting parameters and environmental conditions (e.g. different casting scenarios) to determine relationships between image pixels and corresponding temperature values in different casting scenarios.

[0128] During model validation, the trained machine learning model's performance is evaluated to ensure its accuracy and reliability. This is typically carried out by using a separate temperature dataset (and not the temperature dataset used during training) for each of the different casting scenarios, to validate the model's predictions. Various metrics and techniques, such as mean absolute error (MAE), root mean square error (RMSE), and cross-validation, can be employed to assess the model's performance.

[0129] During prediction and visualization, the trained and validated machine learning model can be used to predict the dynamic temperature distribution in real-time or near realtime for different casting scenarios. In particular, the casting scenarios may account for variations in the relevant casting parameters and environmental conditions, such as casting speed, molten metal temperature, and cooling system settings. The trained and validated model can therefore provide estimations of the temperature distribution throughout the chamber 110. These predictions may be visualized using graphical representations, heatmaps, or animations to aid in understanding the temperature dynamics.

[0130] During optimization and decision-making, the estimated dynamic temperature distribution can be leveraged to optimize the SMA foil fabrication process. For instance, by analysing the predicted temperature profiles, engineers can identify areas of excessive heat or cooling deficiencies, leading to potential defects or suboptimal casting quality. With this information, adjustments may be made to casting parameters, cooling strategies, or system designs to optimize the temperature distribution and enhance the overall process efficiency.

[0131] Moreover, the iterative nature of machine learning methodologies allows continuous learning and improvement. By collecting new temperature data and refining the model, its accuracy and predictive capabilities can be enhanced over time, resulting in better estimations of the dynamic temperature distribution in the planar flow caster 102.

[0132] Typically, one or more machine learning models can be used in the method 200 of generating thermal distribution maps 202, 204, 206. For example, the machine learning models used may include a combination of unsupervised computer vision techniques, unsupervised pixel clustering method and gradient boosting algorithms (e.g. XGBOOST).

[0133] The trained machine learning model in step 220 may be a supervised learning model using gradient boosting algorithms. The supervised gradient boosting model may be trained using generated clusters of pixels, based on mean cluster centroids. For example, in each of the clusters, pixels may be labelled with a number (e.g., 1, 2, 3...). The labelled pixels and pixel clusters are provided as input to the gradient boosting model, and the training targets may be known or estimated a maximum temperature value corresponding to each of the specific zones or clusters. As previously mentioned, temperature datasets may be compiled using temperature sensors strategically mounted throughout the chamber 110.

[0134] Typically, the gradient boosting model may be initially trained with real video footage data samples collected during a 'dry' experiment, without any molten metal involved in the planar flow caster 102. The initially trained gradient boosting model may be tested using real video footage data during a 'wet' experiment, with molten metal flowing during operating of the planar flow caster 102, to predict thermal zones and a maximum temperature value for each of the zones within a video frame dynamically.

[0135] Advantageously, the method 200 may be used to determine dynamic temperature distribution of the chamber 110 or a focused area of interest within the chamber 110. The dynamic temperature distribution provides information in pertaining to the variation of temperature within the chamber 110 or a portion of the chamber 110 over time. Unlike a static temperature distribution, which remains constant, a dynamic temperature distribution considers the changing temperatures at different points within the chamber 110 or a portion of the chamber 110 as time progresses.

[0136] In many practical scenarios, dynamic temperature distribution occurs due to factors such as heat generation, heat transfer, convection, radiation, and other thermal processes. Understanding the dynamic temperature distribution may be beneficial for several reasons. For example, it may facilitate assessment of the thermal performance of system 100, predicting potential overheating issues, optimizing cooling strategies, and ensuring the safe operation of the planar flow caster 102. Moreover, by monitoring and analysing the temperature distribution, engineers can make informed decisions about design improvements, material selection, and thermal management techniques. As such, embodiments of the present invention may facilitate optimisation of the planar casting process, improve SMA foil product quality, and enhance energy efficiency.Variable distance between melt distribution receptacle and the cooling surface

[0137] Figures 3A to 3C further illustrate a setup of the heating coil 108, melt distribution receptacle 112 and the casting wheel 114 having cooling surface 116. During operation of the planar flow caster 102, the melt distribution receptacle 112 gradually moves towards the casting wheel 114. Figure 3A shows the setup at an initial operating stage in which a separation between the melt distribution receptacle 112 and the cooling surface 116 may be about 10mm. As illustrated in Figure 3B, a reflection 134 of the melt distribution receptacle 112 appears on the casting wheel 114 as the melt distribution receptacle 112 moves towards the cooling surface 116. As further illustrated in Figure 3C, the size of the reflection 134 increases the closer the melt distribution receptacle 112 moves towards the cooling surface 116. During operation, the molten metal typically starts to flow from the melt distribution receptacle 112 towards the cooling surface 116 only after the distance between the melt distribution receptacle 112 and the cooling surface 116 is less than about 2mm(Figure 3C). As described in further detail below, embodiments of the present invention use the position of the melt distribution receptacle 112 and its reflection 134 in the visual data to determine the variable distance between the melt distribution receptacle 112 and the cooling surface 116.

[0138] As shown in Figure 3A, physical measurements (pm) marked as 'pml' to 'pm4' related to melt distribution receptacle 112 (pml), the variable distance between the melt distribution receptacle 112 and the cooling surface 116 (pm2), the casting wheel 114 (pm3), and the heating coil 108 (pm4) respectively may be used as reference measurements to calibrate the video data from the camera 118. In particular, these physical measurements may be used to calibrate the captured video frames to predict and estimate dynamic and variable distance 136 and flow velocity of the molten metal (Figure 4A) in real units as described in further detail below. In embodiments where the focused area of interest 228 for thermal distribution is consistent with the components shown in Figures 3A to 3C, the visual data used to generate the thermal distribution maps as described above in relation to method 200 may also be calibrated using the physical measurements 'pml' to 'pm4' as shown in Figure 3A.

[0139] A method 300 of determining a variable distance 136 between the melt distribution receptacle 112 and the cooling surface 116 will now be described with reference to Figure 3D.

[0140] At step 302, video data 350 is pre-processed into a series of image frames (e.g. coloured RGB image frames).

[0141] At step 304, the pre-processed image frames are converted to greyscale image frames. The pixels in each greyscale image frame typically have intensity values (or grey levels) of 0 to 255.

[0142] At step 306, a dynamic thermal threshold is applied in a similar manner as previously described with reference to method 200 in step 214.

[0143] At step 308, bright connected pixels within each greyscale image frame are detected in a similar manner as previously described with reference to method 200 in step 216.

[0144] At step 310, connected components or areas within each greyscale image frame are located, for example using connected component analysis.

[0145] At step 312, contour boundaries of the connected components or areas detected in step 310 are determined, for example using contour tracing algorithms.

[0146] At step 314, an area size of the connected components within each contour boundary is calculated.

[0147] At step 316, two components within each image frame having the largest area size as calculated in step 314 are identified. The two identified components are considered to represent the melt distribution receptacle 112 and its reflection 134, and are therefore denoted as two regions of interest (ROI) for each image frame. As illustrated in Figures 3B and 3C, determining a distance between the melt distribution receptacle 112 and its reflection 134 determines the variable distance between the melt distribution receptacle 112 and the cooling surface 116.

[0148] At step 318, the respective centroids of the two ROIs as defined by the contour boundaries are determined.

[0149] At step 320, bounding box annotation or localization of each of the two ROIs is performed by delineating or outlining a rectangular region that encloses each ROI. This is illustrated in image frame 328 in Figure 3D. Rectangle A 330 outlines a first ROI corresponding to the melt distribution receptacle 112 and rectangle B 334 outlines a second ROI corresponding to the reflection 134. Rectangle A 330 may define a first boundary corresponding to the melt distribution receptacle 112. Rectangle B 334 may define a second boundary corresponding to the cooling surface 116.

[0150] In some embodiments, the rectangle A 330 and rectangle B 334 may be determined using one or more unsupervised machine learning models. In particular, a combination of unsupervised computer vision techniques, and unsupervised pixel contour detection methods may be used.

[0151] At step 322, a third rectangle (rectangle C 332) is generated enclosing the area between rectangle A 330 and rectangle B 334 as shown in image frame 328. A height ofrectangle C 332 representing a distance between the first boundary and the second boundary enables determination of the variable distance 136.

[0152] At step 324, the variable distance 136 between the first boundary and the second boundary is calculated by determining the height of rectangle C in each image frame.

[0153] In some embodiments, the distance 136 between the first boundary and the second boundary may be determined using a supervised machine learning model. In particular, a layer of supervised gradient boosting algorithm (e.g. XGBOOST) may be employed to predict the actual values of the variable distance 136.

[0154] The supervised gradient boosting algorithm may be trained with real video footage data samples collected during a 'dry' experiment, whilst operating the planar flow caster 102 without using any molten metal. Testing of the trained model may be conducted using real video footage data during a 'wet' experiment, whilst operating the planar flow caster 102 with molten metal flowing.

[0155] In some embodiments, the distance 136 is determined by first vertically estimating the gradually increasing area size covered by vertically positioned black masked pixels as shown in the sequence of images 336, representing downwards moving melt distribution receptacle 112 and its reflection 134. The 'gradually decreasing gap' between the receptacle 112 and reflection 134 is the variable distance 136. During operation, the molten metal flows between this gap.

[0156] At step 326, the processed image frames illustrating the reduction in the variable distance 136 can be visualised for example as shown in image frames 336.

[0157] Steps 304 to 326 of method 300 is repeated until all image frames of a complete video sequence are processed.

[0158] In a planar flow caster 102, the variable distance 136 is an important operating parameter that affects the casting process and the quality of the cast SMA foil product. For instance, the variable distance 136 may affect the molten metal flow, cooling and solidification of the molten metal, defect formation in the SMA foil product, heat transfer, and process control and optimisation.

[0159] Elaborating further, the variable distance 136 influences the flow of molten metal from the melt distribution receptacle 112 onto the cooling surface 116. A smaller distance 136 restricts the flow of molten metal, resulting in reduced metal throughput and potential casting defects. On the other hand, a larger distance 136 can lead to excessive metal flow, which may cause instability, uneven thickness, or splattering.

[0160] Moreover, the variable distance 136 affects the cooling and solidification process of the molten metal as it moves across the cooling surface 116 of the spinning casting wheel 114. A smaller distance 136 allows for more rapid cooling due to the reduced distance for heat transfer, resulting in quicker solidification. Conversely, a greater distance 136 provides more time for cooling, potentially allowing for larger grain structures or changes in the microstructure of the SMA foil product.

[0161] Furthermore, improper settings for the distance 136 can contribute to various defects in the SMA foil product. A very small distance 136 can lead to incomplete filling of the mould or insufficient contact between the molten metal and the cooling surface 116, resulting in incomplete castings or weak bonds. On the other hand, a large distance 136 can cause excessive turbulence and air entrapment, leading to porosity, voids, or surface imperfections.

[0162] In addition, the variable distance 136 affects heat transfer between the spinning casting wheel 114 and the molten metal. A smaller distance 136 promotes better heat transfer, allowing for efficient cooling and solidification. Conversely, a greater distance 136 can reduce the heat transfer rate, potentially leading to longer solidification times, thermal gradients, and non-uniform properties within the cast SMA foil material.

[0163] Controlling and adjusting the variable distance 136 also facilitates achieving desired casting properties and quality. Optimisation of the distance 136 involves considering various factors such as the type of material being cast, its melting characteristics, desired cooling rates, and the overall casting process parameters. It often requires a balance between molten metal flow, solidification time, cooling requirements, and the desired mechanical and microstructural properties of the final SMA foil product.

[0164] To determine the optimal distance 136, experimental trials, empirical models, and simulation techniques can be employed. These methods consider the specific requirements of the casting process and the desired quality criteria to establish appropriate settings for the variable distance 136.

[0165] Applying machine learning algorithms to estimate and monitor the variable distance 136 in accordance with embodiments of the invention, may be beneficial in achieving process improvement and optimisation in a number of ways. For example, the determination of the optimal distance 136 involves considering various parameters, such as molten metal properties, casting conditions, cooling requirements, and desired casting quality. The relationship between these factors and the ideal distance 136 setting can be highly complex and nonlinear. Machine learning algorithms such as those described herein may be particular suited for determining the complex and nonlinear relationships between these factors.Flow velocity of molten metal

[0166] A method 400 for determining a dynamic flow velocity of the molten metal from the melt distribution receptacle 112 towards the cooling surface 116 will now be explained with reference to Figures 4A to 4C. As described in further detail below with specific reference to Figure 4A, method 400 for determining the dynamic flow velocity is interlinked with the method 300 for determining the variable distance 136.

[0167] Generally, method 400 follows similar steps to those described above in method 300 until the variable distance 136 is less than or equal to 2mm. After this stage, the method 400 estimates an effective diameter in use and an effective length of molten metal flowed at a given time, using the number of black pixels present within a defined region of interest 414 (rectangle D) between the melt distribution receptacle 112 and the cooling surface 116 as explained in further detail below.

[0168] In method 400, like references refer to those previously described. In particular, steps 302 to 320 are the same as those described above with reference to Figure 3D in method 300.

[0169] At step 402, a region of interest 414 marked by rectangle D is located within each image frame when the distance 136 is less than or equal to about 2mm.

[0170] At step 404, a total number of dark pixels (e.g. pixels having a grey level of 255) extending across a width direction 426 of rectangle D 414 is estimated.

[0171] At step 406, a percentage of 'used width' is calculated based on the total number of dark pixels estimated in step 404 and a width of rectangle D. As the width of the rectangle D in 'mm' can be estimated using physical measurement 'pm4' (see Figure 3A). This percentage may be used to determine the effective diameter 422 in mm of the molten metal as illustrated in image frame 418.

[0172] At step 408, a total number of dark pixels (e.g. pixels having a grey level of about 255) extending across a height direction 428 of rectangle D 414 is estimated.

[0173] At step 410, a percentage of 'used height width' is calculated based on the total number of dark pixels estimated in step 408 and a height of rectangle D. As the height of the rectangle D in mm can be estimated using physical measurement 'pm2' (see Figure 3A), this percentage may be used to determine the effective height 424 in mm of the molten metal as illustrated in image frame 418.

[0174] At step 412, the flow velocity of the molten metal may be calculated in mm3 / seconds based on the effective diameter 422 (effective cross-sectional area) of the molten metal and the effective height 424 representing the distance over which the molten metal has flowed.

[0175] In some embodiments, flow velocity may be determined using a supervised machine learning model. In particular, an additional layer of supervised gradient boosting algorithm (e.g. XGBOOST) may be employed to predict the actual values of the dynamic flow velocity of the molten metal, for example in mm3 / seconds.

[0176] Typically, the above steps of method 400 are repeated for each image frame of the complete video sequence captured by the camera 118.

[0177] Experimental results for the variable distance 136 as determined using method 300 are illustrated in Figure 4B. In Figure 4B, the horizontal x-axis represents time in seconds, and the vertical y-axis represents distance in mm.

[0178] Similarly, experimental results for the flow velocity as determined using method 400 are illustrated in Figure 4C. In Figure 4C, the horizontal x-axis represents time in seconds, and the vertical y-axis represents velocity in m3 / s.

[0179] As shown in Figures 4B and 4C, as the variable distance 136 reduces, the flow velocity of the molten metal increases.

[0180] Measuring the flow velocity of molten metal during operation of a planar flow caster 102 beneficially provides information about the casting process which may affects the quality and properties of the cast product. In particular, the flow velocity may be used in process optimisation, defect detection and prevention, uniformity and thickness control, cooling and solidification process refinement, process monitoring and control, simulation and modelling.

[0181] For example, the flow velocity of molten metal may influence the casting process parameters and can impact the overall efficiency and quality of the castings. By accurately measuring and monitoring the flow velocity, engineers can optimize process conditions, such as casting speed, to achieve desired cooling rates, solidification behaviour, and control over the microstructure of the cast SMA foil material.

[0182] Moreover, measuring flow velocity may facilitate identification of potential defects in the casting process. Flow disturbances, turbulence, or uneven flow velocities may result in defects like porosity, inclusions, or uneven solidification. By monitoring flow velocities, any deviations or irregularities can be detected early, allowing for corrective measures to be taken to prevent or minimize the occurrence of defects.

[0183] Furthermore, the flow velocity may affect the distribution of molten metal on the casting wheel 114. By measuring flow velocity, engineers can ensure a uniform and controlled flow across the entire cooling surface 116, leading to more consistent thickness and improved dimensional accuracy of the cast SMA foil product.

[0184] Additionally, the flow velocity may be used for controlling the cooling and solidification process. A higher flow velocity can enhance heat transfer, resulting in faster cooling and shorter solidification times. Conversely, a lower flow velocity allows for more controlled heat dissipation, promoting longer solidification times. By measuring flow velocity, engineers can optimize cooling rates and control the solidification behaviour to achieve desired material properties and casting quality.

[0185] Flow velocity measurements may also provide real-time feedback on the status of the casting process. By continuously monitoring flow velocities, engineers can detect any deviations from the desired range, promptly identify process anomalies, and adjust as necessary. This capability enables better process control, reduces the risk of defects, and improves the overall consistency of the cast SMA foil product.

[0186] Accurate flow velocity measurements may serve as valuable inputs for numerical simulations and modelling of the casting process. Computational Fluid Dynamics (CFD) models can utilize flow velocity data to simulate and predict the flow behaviour, temperature distribution, and solidification patterns within the planar flow caster 112. These simulations aid in process design, optimization, and prediction of casting quality.

[0187] As such, measuring flow velocity in a planar flow caster 102 may be beneficially used for process optimization, defect detection and prevention, uniformity control, cooling and solidification management, process monitoring and control, as well as simulation and modelling. By accurately measuring and understanding the flow velocity of molten metal, engineers can make informed decisions to enhance the casting process, improve product quality, and achieve desired material properties.

[0188] Using computer vison, image processing and machine learning techniques to determine the flow velocity of molten metal in accordance with embodiments of the present invention may improve existing planar flow casting processes. For example, traditional methods of measuring flow velocity often involve intrusive techniques, such as using probes or physical sensors, which can interfere with the casting process and affect the accuracy of measurements. Embodiments of the present invention, on the other hand, provides nonintrusive solutions, relying on data collected from various sensors or cameras placedstrategically within the chamber 110. This non-intrusive approach allows for continuous and real-time monitoring of flow velocity without disrupting the casting process.

[0189] Further, determining dynamic flow velocity in accordance with embodiments of the present invention enables real-time monitoring and control of the casting process. The dynamic flow velocity may be used for real-time adjustments, ensuring that the flow velocities remain within the desired range and allowing for timely interventions to optimize the casting process and prevent defects.

[0190] The machine learning model used for determining flow velocities may be trained with a diverse dataset that includes various process parameters and corresponding flow velocities. The trained model may capture intricate relationships and provide accurate and precise flow velocity estimations. This improved accuracy may enhance the overall control of the casting process and reduces uncertainties associated with conventional flow velocity measurements.

[0191] By analysing the estimated flow velocities in relation to other process parameters, such as casting speed, cooling conditions, and mould design, engineers can identify optimal settings that promote uniform flow, controlled solidification, and desired material properties. This optimization leads to improved casting quality, reduced defects, and enhanced overall process efficiency.

[0192] The machine learning models can learn from historical data to identify patterns that may not be readily apparent to human operators. By analysing the estimated flow velocities, engineers can gain valuable insights into the process dynamics, identify correlations with casting quality, and make data-driven decisions to optimize the process parameters and achieve desired outcomes.

[0193] Additionally, the machine learning models can continuously learn and adapt to changing conditions. As the planar flow caster 102 operates over time, new data can be collected and fed into the machine learning model, allowing it to update and improve its flow velocity estimations. This adaptive capability beneficially allows the machine learning to be up to date with variations in casting conditions, process modifications, and other factors influencing flow velocity, leading to enhanced accuracy and performance over time.

[0194] In summary, estimating flow velocity in accordance with embodiments of the present invention may provide several advantages, including non-intrusive measurement, real-time monitoring and control, improved accuracy and precision, process optimization, data-driven decision making, and continuous learning. These benefits contribute to better process control, enhanced casting quality, and increased efficiency in the SMA foil fabrication process.Control Module

[0195] Now reverting to Figure IB, the system 100 may further include a control module 106, such as a closed-loop control module operatively configured to determine one or more control parameters based on one or more operating parameters (thermal distribution 124, estimated variable distance 126, and flow velocity 128 of the molten metal) determined by the process monitoring module 104. The control parameters may include a temperature of the heating coil 138, a rotating speed of the casting wheel 140, a descending speed of the melt distribution receptacle 142, and a pressure generated during the SMA melting process 142. Other control parameters may be used to control operations of the planar flow casting fabrication unit. These other control parameters may also be adjustable via the control module 106.

[0196] The control parameters 138, 140, 142, 144 may be determined and / or adjusted using a feedback control system. For example, the speed of rotation for the casting wheel 114 may be adjusted in response to the measured flow velocity 128 of the molten metal. In one scenario, if the flow velocity 128 of the molten metal is below a predetermined threshold, the rotational speed 140 of the casting wheel 114 may be proportionally reduced based on a predetermined correlation between optimum flow velocity 128 and rotational speed 140 of the casting wheel 114 for a range of SMA materials. A different correlation may be determined for each type and mass of SMA material used.

[0197] In another example, when the estimated flow velocity 128 of the molten is generally consistent between process runs, irregular and / or unexpected temperature fluctuations within the chamber 110 as detected via the thermal distribution maps 124 may cause the fabricated SMA foil product to be too thin, resulting in a failed foil product. A failed foil product generally indicates that one or more of the control settings 138, 140, 142, 144 ofthe control module 106 requires adjustment. In this scenario, the control module 106 may adjust (increase or decrease) the pressure generated during the SMA melting process 142 as required. The adjustment of the pressure 142 may correspondingly adjust the speed and / or flow behaviour of the molten metal onto the casting wheel 114 to alter the thickness of the final SMA foil product. Alternatively, or in combination, the speed 144 of the casting wheel 114 may be adjusted by the control module 106 to further adjust the thickness of the SMA foil product.

[0198] In some embodiments, the control module 106 may be used to adjust other control parameters in addition to those shown in Figure IB. For example, the estimated thermal distribution 124 may be used to determine an optimal position of the melt distribution receptacle 112 relative to the heating coil 108. In practice, the optimal position of the melt distribution receptacle 112 relative to the heating coil 108 may change with changes in weather and other environmental conditions, fluctuations in the power supply, contamination in the raw material for the molten metal, and / or variations in magnetic induction based on the design of the heating coil 108. In some instances, it may be desirable for the melt distribution receptacle 112 (i.e. a melting position of the molten metal) to be generally aligned with a hottest zone in the temperature distribution of the chamber 110. As such, if the thermal distribution maps 124 indicate that he melt distribution receptacle 112 is lower than the hottest zone in the temperature distribution of the chamber 110, the relative position of the melt distribution receptacle 112 may be adjusted (e.g. via a pulley wheel or similar adjustment mechanism to adjust the mechanical distance between the melt distribution receptacle 112 and casting wheel 114) such that the melt distribution receptacle 112 is generally aligned with the hottest zone as indicated by the thermal distribution 124.

[0199] Advantageously, the control module 106 maybe capable of determining optimised control parameters 138, 140, 142, 144 to optimise operations of the planar flow caster 102 so as to optimise the SMA foil fabrication process to provide an SMA foil product of improved quality. This may be automated, thereby reducing the need for manual calibration of the planar flow caster, and trial and error, which can be time consuming and costly as it is likely to lead to wastage of the raw materials. Moreover, the monitoring and control performed by monitoring module 104 and the control module 106 may be achieved in real-time or near real-time, thereby enabling instantaneous adjustment of the controlparameters based on the operations of the planar flow caster 102. This may allow a more precision based fabrication environment to produce an SMA foil product of a desired consistency and quality. As such, SMA foil products produced via the SMA foil fabrication process of system 100 may be capable of achieving a width of about 30mm to 100mm.Amorphous Metal / Alloy Foil Fabrication

[0200] Amorphous metal foil (also referred to herein as amorphous alloy foil) is a material used in applications in which characteristics such as high strength, corrosion resistance, and suitable magnetic properties are desirable. Whilst the specification may refer to metal foil and alloy foil interchangeably, it will be understood that embodiments of the invention may apply to any one or both of metal and alloy foil fabrication, and reference to either one does not limit embodiments of the invention to the fabrication of a metal or alloy foil.

[0201] Unlike shape memory alloy (SMA), amorphous metal lacks crystallinity. To maintain the non-crystalline structure of amorphous metal during the fabrication of amorphous metal foil by planar flow casting, it is desirable to achieve rapid cooling which requires stringent control over cooling rates and flow dynamics during the planar flow casting process. Furthermore, precise monitoring and controlling of operating parameters are useful for maintaining the desired mechanical and magnetic properties of the amorphous metal foil, whilst minimising defects such as cracks, voids and uneven thickness.

[0202] Embodiments of the present invention provide methods for monitoring and controlling one or more operating parameters of planar flow casting for amorphous metal foil fabrication. By utilising monitoring and control systems such as real-time monitoring and control systems, including advanced video analysis algorithms and machine learning models, embodiments of the present invention may enhance the efficiency and accuracy of the fabrication of amorphous metal foil by planar flow casting.

[0203] With reference to Figure 5A, there is illustrated a planar flow caster 500 suitable for fabricating amorphous metal or alloy foils. The planar flow caster 500 includes a high temperature chamber 501. The temperature of the chamber may be maintained at more than 1500°C. The planar flow caster 500 further includes a heating coil 502 for heating a massof metal to produce molten metal. The molten metal is extruded through a molten metal receptacle. Typically, the molten metal receptacle comprises a crucible 503 coupled with a nozzle. The planar flow caster 500 further includes a molten metal flow slider 504, and a molten metal collector 506. During operation, molten metal flows from the crucible 503 onto the molten metal flow slider 504 through the nozzle. The molten metal subsequently slides onto a molten metal collector 506. After the molten metal leaves the molten metal collector 506, it flows onto a spinning casting wheel 508 providing a cooling surface which rapidly cools the metal and produces amorphous metal foil 510. The planar flow caster 500 further includes a camera (not shown) operatively configured to capture visual data of any or all parts of the melt distribution receptacle, the flow path of the molten metal between the melt distribution receptacle 503 and the casting wheel 508, and the casting wheel 508 during operation. The planar flow caster 500 may also comprise additional elements, analogous to the planar flow caster 100 described previously with reference to Figure IB.

[0204] As it will be appreciated, precise process monitoring and control in the planar flow casting process may allow a consistent molten flow onto the casting wheel. This may advantageously improve the uniformity, surface quality, mechanical integrity and magnetic properties of the amorphous metal foil and improve process control and manufacturing efficiency. As it will be understood, the tolerance in thickness and uniformity may be especially low for amorphous metal foils in some circumstances, as they may significantly alter the mechanical, electrical and magnetic properties of the amorphous metal foil.

[0205] Figure 5B provides a high-level summary of a system for implementing methods 600 and 700 of calculating the operating parameters outlined below. In this system, video frames of molten metal flow are captured by video camera 522 and are subsequently pre- processed by a pre-processing module 524. The pre-processed video frames are then analysed by flow rate calculation module 526 and / or wheel speed measurement module 528 to determine the relevant operating parameter(s) by way of a trained model such as deep learning model 532. The flow rate calculation module 526 is configured to carry out a computer implemented method 600 as described herein with reference to Figure 6A. The wheel speed measurement module 528 is configured to carry out a computer implemented method 700 as described herein with reference to Figure 7.

[0206] Optionally, the system may also comprise a foil dimension analysis module 530 to calculate the dimensions of the foil as a quality parameter, as described herein with reference to Figure 8A. The foil dimensions as well as any other suitable quality parameters may be determined using a trained model, such as a deep leaning model 532. In some embodiments, the operating parameters and / or quality parameters may be transmitted to control system 534 to carry out closed-loop control of the planar flow caster 500. As will be understood, during production, the deep learning model 532 can take the real-time flow rate, wheel speed, and foil dimensions as input, and predict the optimal control settings to maintain the desired quality of the metal foil. Although illustrated as separate modules, it will be understood that modules 524, 526, 528, 530 and 532 may also be integrated into a single module in certain embodiments.

[0207] Methods of calculating operating parameters including flow rate and wheel speed, and methods of calculating one or more quality parameters, such as dimensions of the foil during the PFC process of amorphous metal foil, will now be described in further detail according to embodiments of the invention.Operating parametersFlow rate of molten metal

[0208] A method 600 of calculating the flow rate of molten metal will now be described with reference to Figure 6A. As used herein, "flow rate of molten metal" refers to the volume of molten metal passing through an arbitrary specific cross-section per unit time.

[0209] At step 601, consecutive video frames of molten metal flow are captured by the camera. Typically, the video frames captured by the camera are RGB coloured. Figures 6B and 6C are example image frames extracted from the video. In particular, Figure 6B is an example image frame taken just prior to full interaction between the molten metal and the spinning casting wheel, capturing a pre-contact phase. Figure 6C is one of the subsequent image frames capturing a moment after the molten metal begins interacting with the casting wheel, capturing a more stabilised phase.

[0210] At step 602, the video frames are pre-processed to generate pre-processed video frames. Typically, the frames are converted to greyscale with varying intensity values beingassigned to each pixel. The intensity values may be normalized using standard techniques such as min-max scaling. The frames may be processed using standard methods such as histogram equalization and noise reduction.

[0211] At step 604, regions of motion representing flowing molten metal are defined as regions of interest (ROI). Specifically, regions of motion are distinguishable from stationary components because they are associated with significant changes in pixel intensity between consecutive video frames. Typically, the difference in pixel intensity between consecutive pre-processed frames is calculated for each pixel, and a threshold is applied to these differences to create a binary mask to highlight areas with significant changes in intensity. If the change in pixel intensity exceeds the threshold, the binary mask will highlight those pixels as regions of interest. Typically, a threshold value between 25 and 50 on a scale from 0 to 255 is used. The threshold value or values may be adaptively changed based on detected motion. As an example, Figures 6E and 6F illustrate output of the edge detection for image frames in Figures 6B and 6C respectively. Figure 6G illustrates the subtraction of the image in Figure 6F from the image in Figure 6E showing highlighted regions of interest obtained from Figures 6E and 6F, which correspond with areas of molten metal flow. Similarly, Figure 6D illustrates the subtraction of the image in Figure 6C from Figure 6B. As can be seen from Figure 6G and 6D, the ROI excludes stationary components which would be seen in a typical frame-differenced image such as that shown in Figure 6D.

[0212] At step 606, region-based segmentation is applied to the binary mask generated in step 604 to generate segmented regions that represent isolated molten metal flow. During segmentation, morphological operations may be employed to clean the binary mask, and connected components are labelled and analysed. The segmentation process may include several substeps such as optical flow calculation 6060, thresholding 6062, flow vector analysis 6064, and visualization 6066. Optical flow calculation 6060 may include using known algorithms such as the Lucas-Kanade method and / or the Farneback method. Thresholding 6063 may include using known algorithms such as adaptive thresholding and / or Otsu's method. Flow vector analysis may include calculating the vector length of each pixel in the segmented regions, wherein the direction of the vector represents the angle of molten metal movement, and the magnitude of the vector represents the speed of the movement. Filtering may further be applied by removing vectors whose magnitude fall below a pre-determinedthreshold. Visualization 6066 may include overlaying vectors on frames. The overlaying vector may be colour-coded based on the direction and / or magnitude of the calculated vector. The region size and shape parameters may be adaptively optimised.

[0213] At step 608, the area of the segmented regions generated in step 606 are measured at fixed time intervals to calculate the flow rate of the molten metal. The area of each segmented region may be calculated based on the number of pixels in each region. The volume of molten metal passing through a given cross-sectional area per unit time may therefore be calculated based on the number of pixels passing through the given crosssection per time interval. Each measurement interval may be optimised and may be in the range of 0.1 to 0.5 seconds. The calculated flow rate may be aggregated over a period of time to obtain a mean flow rate. Typically, the calculated flow rate has a unit of cubic centimetres per second (cm3 / s).

[0214] One example flow rate calculation will now be described as follows.

[0215] As mentioned, the flow rate of molten metal onto the spinning casting wheel can be calculated by analysing the volume of metal passing through a specific cross-section per unit time. Using consecutive frames from the video, the area of the molten metal in each frame may be measured to track the movement of the molten metal and to estimate the flow rate.

[0216] As described above in steps 601 to 608 of method 600, image frames are extracted from the video at regular intervals (e.g., 30 frames per second) during a frame extraction process. Edge detection can then be applied to each frame to identify the boundaries of the molten metal flow. For each frame, the area of the metal flow (in pixels) may be calculated. This area represents the volume of metal that passes through a cross- sectional region of the casting setup. Using the region of interest (ROI) determination as described with reference to step 604, the movement of the metal across consecutive frames is tracked to determine the distance the molten metal has moved.

[0217] As such, a formula for determining a flow rate of molten metal may be:Flow Rate (cm3 / s) = (Area of Metal Flow * Distance Moved) / Time Interval Between Frames

[0218] In one example calculation:Area of Metal Flow (Figure 6E: Frame 1): 500 pixels2.Area of Metal Flow (Figure 6F: Frame 2): 520 pixels2(due to slight expansion).Distance Moved: 10 pixels.Time Interval between Frames 1 and 2: 1 / 30 sec (for 30 FPS).Flow Rate = (500 + 520) / 2 * 10 / (1 / 30) = 15,300 pixels3 / sec.Speed of casting wheel

[0219] A method 700 for determining the speed of the spinning wheel 508 will now be described with reference to Figure 7.

[0220] At step 702, video frames of the spinning casting wheel 508 are captured by the camera. Typically, the video frames captured by the camera are RGB coloured. The video frames may capture the entirety of the spinning casting wheel or a part of the wheel. Figures 6B and 6C respectively illustrate part of the spinning casting wheel 508 in a pre-contact phase, and a more established phase with a steady flow of molten metal, as captured by the camera.

[0221] At step 704, the video frames are pre-processed to generate pre-processed video frames. Typically, the frames are converted to greyscale with varying intensity values being assigned to each pixel. The intensity values may be normalized using standard techniques such as min-max scaling. The frames may be processed using standard methods such as histogram equalization and noise reduction.

[0222] At step 706, a binary edge map is generated from the pre-processed video frames using gradient-based methods. As an example, the Canny Edge Detector may be used. The binary edge map highlights pixels with sharp changes in intensity values and is generated by a thresholding process. If the change in pixel intensity exceeds a threshold, the binary edge map will highlight those pixels as edge of the spinning wheel 508. As an example, binary edge maps generated from Figures 6B and 6C are shown in Figures 6E and 6F, respectively.

[0223] At step 708, the spinning wheel 508 is identified using shape detection algorithms. The step 7080 may include several substeps such as Hough Transform for circles 7080, circle parameter extraction 7082, and visualization 7084. Typically, in the Hough Transform substep 7080, edge points are extracted from the binary edge map and are used to vote for possible circles in a three-dimensional parameter space defined by the circle's centre coordinates and radius. Each edge point votes for all potential circles that could pass through it, and these votes are accumulated in an array, which may be referred to as an accumulator array. Peaks in this accumulator array indicate the most likely circles. Parameters of the circles, such as the x-y coordinates of the circle centre and the radius, are subsequently extracted at substep 7082. At substep 7084, the circles may be visualized via an overlay on the video frames. The visualization may be colour coded based on the radius. It will be appreciated that other methods may also be used to identify the spinning wheel 508.

[0224] At step 710, a linear wheel speed in metres per second (m / s) is obtained based on the angular velocity of the wheel and the radius of the wheel (which is typically known). The angular velocity estimation process may include the substeps of feature detection 7100, feature tracking 7102, angular displacement calculation 7104, angular velocity estimation 7106, and visualization 7108. Typically, at substep 7100, features of interest on the spinning wheel may be detected using known algorithms such as Harris Corner detection. In the case of Harris Corner detection, pixels corresponding to points where the intensity changes significantly in multiple directions are identified. At substep 7102, the features of interest may be tracked using standard optical flow methods such as the Lucas-Kanade algorithm, which estimates motion by analysing changes in pixel intensity. At substep7104, the angular displacement of the tracked features of interest are calculated based on the change in angle between consecutive positions, as captured in consecutive image frames. At substep 7106, angular velocity of the wheel is estimated by calculating the change in angular displacement of tracked features over time. The estimated angular velocity may optionally be visualized via an overlay on the frames at substep 7108. The visualization may be colour coded based on the magnitude of the estimated angular velocity. As it will be appreciated, the speed of the wheel may also be expressed in revolutions per minute (RPM).

[0225] One example casting wheel speed calculation will now be described below.

[0226] As described in further detail below, the rotational speed of the casting wheel may be used in determining the dimensions (e.g. final thickness) of the amorphous metal foil. In some embodiments, the wheel speed can be measured by analysing the angular motion of the wheel between consecutive frames.

[0227] As previously mentioned, frames are extracted from the video where the wheel is clearly visible. The casting wheel may be identified using a Hough Circle Transform to detect the circular edges of the wheel. By tracking distinct features on the wheel (such as surface markings or spokes), the angular displacement of the wheel between frames may be measured.

[0228] As such, a formula for determining a flow rate of molten metal may beWheel Speed (m / s) = (Angular Displacement (radians)) / (Time Interval Between Frames) * Wheel Radius (m)

[0229] In one example calculation:Angular Displacement: 0.1 radians.Wheel Radius: 0.5 meters.Time Interval: 1 / 30 sec.Wheel Speed = (0.1) / (l / 30) * 0.5 = 1.5 m / s.Quality Parameters

[0230] As mentioned earlier, in some cases, it may be desirable to dynamically monitor various quality parameters during the casting flow process to provide timely feedback on the quality of the metal foil as it is being fabricated on the spinning casting wheel. Such quality parameters may include dimensions (e.g. width and / or thickness) of the metal foil as described in more detail below with reference to Figure 8A, and percentage variation in foil thickness across the width as described in more detail below with reference to Figures 8B to 8E. For example, for certain applications, such as those in aerospace engineering, it is often desirable to obtain metal foils of a high degree of uniformity in thickness across its width toachieve certain material characteristics such as high reflectivity. Similarly, certain applications require metal foils of specific dimensions, particularly in electronics and sensor applications. Accordingly, metal foils with a percentage variation in thickness and / or thickness and / or width outside an acceptable tolerance may need to be discarded.

[0231] Embodiments of the present invention may provide real-time or near real-time feedback on the quality of the metal foil as it is being fabricated. This may minimise the production of defective materials, enhance quality control, improve production consistency and efficiency, and lead to cost savings.

[0232] A method of determining dimensions of the foil will be described with reference to Figure 8A. The dynamic monitoring of percentage variation of thickness across the width of the metal foil will be broadly described in relation to Figures 8B to 8E.Dimensions of foil

[0233] A method 800 for determining and monitoring the dimensions of the fabricated amorphous metal foil will now be described with reference to the flow diagram shown in Figure 8A.

[0234] At step 802, video frames of amorphous metal foil 510 are captured by the camera. Typically, the video frames captured by the camera are RGB coloured. Figures 6B and 6C illustrate the amorphous metal foil 510 as it solidifies on the spinning casting wheel 508.

[0235] At step 804, the video frames are pre-processed to generate pre-processed video frames. Typically, the frames are converted to greyscale with varying intensity values being assigned to each pixel. The intensity values may be normalized using standard techniques such as min-max Scaling. The frames may be processed using standard methods such as histogram equalization and noise reduction.

[0236] At step 806, a binary edge map is generated from the pre-processed video frames using gradient-based methods (e.g. see Figures 6E and 6F). As an example, the Canny Edge Detector may be used. The binary edge map highlights pixels with sharp changes in intensity values and is generated by a thresholding process. If the change in pixel intensity exceeds a threshold, the binary edge map will highlight those pixels as foil edge. Typically, an edgethreshold value between 50 and 150 on a scale from 0 to 255 is used for the edge detection. Gaussian smoothing may be applied to the binary edge map to reduce noise. As an example, binary edge maps of the amorphous metal foil 510 generated from Figures 6B and 6C are shown in Figures 6E and 6F, respectively.

[0237] At step 808, texture descriptors are generated from the pre-processed video frames of the amorphous metal foil surface. Typically, the texture descriptors characterise surface texture features such as contrast, correlation, energy and homogeneity. These texture features may be extracted from spatial variations in pixel intensities using known techniques such as Gabor Filters and Gray Level Co-occurrence Matrix (GLCM). Relevant parameters of the Gabor Filters and the GLCM may be optimised.

[0238] At step 810, physical dimensions of the amorphous metal foil are calculated based on the binary edge map and texture descriptors generated in steps 806 and 808, respectively. Specifically, the width of the foil may be measurement by counting the number of pixels between the leftmost and rightmost edges. The thickness of the foil may be directly determined using texture analysis. As will be appreciated, edge detection and texture analysis may be correlated.

[0239] One example foil dimension calculation will now be described below.

[0240] The width and thickness of the amorphous metal foil are useful parameters that may be monitored throughout the PFC process. As mentioned, edge detection and region segmentation techniques may be used to measure these dimensions, for example in realtime.

[0241] As mentioned, the edges of the metal foil may be detected using the Canny edge detection method. The width of the foil may be measured by detecting the distance between the leftmost and rightmost edges of the foil in each frame, and then converted into real- world units. The thickness of the foil may be measured by tracking a vertical spread of the foil edges over time.

[0242] In one embodiment, formulas that may be used to measure the metal foil dimensions may be set out as follows:Width (cm) = (Right Edge Position - Left Edge Position) / Scaling FactorThickness (cm) = (Upper Edge Position - Lower Edge Position) / Scaling Factor

[0243] In one example calculation:Width in Pixels (Figure 6E: Frame 1): 200 pixels.Width in Pixels (Figure 6F: Frame 2): 202 pixels.Scaling Factor: 20 pixels / cm.Width (cm) = 200 / 20 = 10 cm.Foil thickness variation

[0244] Now referring to Figures 8B to 8E, a method for determining the foil thickness variation will be described. Figure 8B is an example image frame taken just prior to full interaction between the molten metal and the spinning casting wheel, capturing a precontact phase. Figure 8C to 8E are each subsequent image frames each capturing a moment after the molten metal begins interacting with the casting wheel, capturing a more stabilised phase. The images are pre-processed to generate greyscale intensity maps, and a region of interest may be determined using similar methods to those described above in relation to method 800. For determining foil thickness variation, the region of interest is typically a portion of the video frame where the foil flows over the spinning wheel, as indicated by the dashed box in Figure 8C. Optionally, contrast in the region of interest may be enhanced using standard image processing methods and pixel intensity thresholding may be used to determine regions that represent the foil. Subsequently, the thickness of the foil may be determined by analysing the intensity of pixels along the width of the foil, where lower intensity values correspond with thinner foil. By analysing the intensity profile of the foil across its width, it is thus possible to determine the percentage variation of the foil. Optionally, the thickness of the foil over time may be visualised using gradient-based methods such as by using a heatmap in which varying thicknesses are represented by a gradient of colours. The heatmap may be optionally overlaid with the image frames, as shown in Figure 8D. Furthermore and with reference to Figure 8E, the image frames may beoverlaid with a heatmap that provides a dynamic visualisation of both the width and thickness variation of the foil over time. As will be understood, in some embodiments, the foil thickness variation may optionally be automatically determined by the system.Feedback control for PFC process of amorphous metal foils

[0245] In some embodiments, experimental data and calculations of the operating parameters such as flow rate of the molten metal, rotating speed of the casting wheel and dimensions of the metal foil as previously described for the relevant input frames may be used to train one or more statistical / machine learning models, such as supervised machine learning models. In one embodiment, data inputs for training a supervised machine learning model may include:• measurements of flow rate, wheel speed, and foil dimensions, for example as determined via vision processing methods described herein, which may be in realtime or near real-time• actual temperature, wheel speed, pressure, and melt distribution speed used during experimental PFC processes• the quality of the resulting foil, as measured by its dimensions, uniformity, percentage variation in thickness across the width of the foil, and other quality control checks.

[0246] Using this dataset, a deep learning model can learn complex relationships between the input parameters and the final product quality. During production, the model can take the real-time flow rate, wheel speed, and foil dimensions as input, and predict the optimal control settings to maintain the desired quality.

[0247] In one embodiment, during operation, a trained model may receive operating parameter data in relation to flow rate (e.g. 15,000 pixels3 / sec), wheel speed (e.g. 1.5 m / s), and foil width (e.g. 10 cm), for example as determined using vision processing methods of determining operating parameters of the PFC process as previously described. In some embodiments, the operating parameters may be determined via statistical / machine learning models in combination with or alternatively to vision processing techniques.

[0248] In some embodiments, based on the input operating parameter data, the trained model may predict that to maintain desirable quality of the final metal foil product, the temperature of the heating coil should be set to 1200°C, the wheel speed should be increased by 2%, and the melt distribution speed should be reduced by 1 mm / s. A control module associated with the planar flow caster may dynamically adjust the relevant control parameters of the planar flow caster to maintain stability and consistency in the casting process and quality of the output amorphous metal foil product.

[0249] In some embodiments, the determined operating parameters for flow rate, wheel speed, and foil dimensions may be used as input parameters for a trained supervised deep learning model to determine any one or more of the following control parameters of a planar flow caster:• Temperature of the heating coil,• Rotating speed of the casting wheel,• Descending speed of the melt distribution receptacle, and• Pressure generated during the metal melting process.

[0250] Embodiments of the invention may therefore leverage deep learning models to provide methods of closed-loop control for the planar flow casing process, thereby providing dynamic adaptation, optimisation, and fault prediction, enabling the casting system to operate at a higher efficiency while maintaining high-quality foil production.Interpretation

[0251] This specification, including the claims, is intended to be interpreted as follows:

[0252] Embodiments or examples described in the specification are intended to be illustrative of the invention, without limiting the scope thereof. The invention is capable of being practised with various modifications and additions as will readily occur to those skilled in the art. Accordingly, it is to be understood that the scope of the invention is not to be limited to the exact construction and operation described or illustrated, but only by the following claims.

[0253] Moreover, any feature or element described within one embodiment may be combined with any feature or element as described with respect to any other embodiment detailed within this specification, as deemed suitable and appropriate by those skilled in the art.

[0254] The mere disclosure of a method step or product element in the specification should not be construed as being essential to the invention claimed herein, except where it is either expressly stated to be so or expressly recited in a claim.

[0255] The terms in the claims have the broadest scope of meaning they would have been given by a person of ordinary skill in the art as of the relevant date.

[0256] The terms "a" and "an" mean "one or more", unless expressly specified otherwise.

[0257] Neither the title nor the abstract of the present application is to be taken as limiting in any way as the scope of the claimed invention.

[0258] Where the preamble of a claim recites a purpose, benefit or possible use of the claimed invention, it does not limit the claimed invention to having only that purpose, benefit or possible use.

[0259] It should be noted that terms of degree such as "generally", "substantially", "about" and "approximately" as used herein mean a reasonable amount of deviation of the modified term such that the end result is not significantly changed. These terms of degree should be construed as including a deviation of the modified term if this deviation would not negate the meaning of the term it modifies.

[0260] In the specification, including the claims, the term "comprise", and variants of that term such as "comprises" or "comprising", are used to mean "including but not limited to", unless expressly specified otherwise, or unless in the context or usage an exclusive interpretation of the term is required.

[0261] Furthermore, the recitation of any numerical ranges by endpoints herein includes all numbers and fractions subsumed within that range (e.g. 1 to 5 includes 1, 1.5, 2, 2.75, 3,3.90, 4, and 5). It is also to be understood that all numbers and fractions thereof are presumed to be modified by the term "about" which means a variation up to a certain amount of the number to which reference is being made if the end result is not significantly changed.

[0262] As used herein, the wording "and / or" is intended to represent an inclusive-or. That is, "X and / or Y" is intended to mean X or Y or both, for example. As a further example, "X, Y, and / or Z" is intended to mean X or Y or Z or any combination thereof.

[0263] Throughout the specification, like reference numerals refer to like features described herein. As such, any instance where features or components are indicated with the same references implies a direct correlation to the similar or identical features or components as previously described in the specification.

[0264] The disclosure of any document referred to herein is incorporated by reference into this patent application as part of the present disclosure, but only for purposes of written description and enablement and should in no way be used to limit, define, or otherwise construe any term of the present application where the present application, without such incorporation by reference, would not have failed to provide an ascertainable meaning. Any incorporation by reference does not, in and of itself, constitute any endorsement or ratification of any statement, opinion or argument contained in any incorporated document.

Claims

The claims defining the invention are as follows:

1. A computer-implemented method for determining operating parameters of a planar flow caster in a metal foil fabrication process, the planar flow caster having a melt distribution receptacle through which a molten metal is extruded, and a cooling surface for cooling and solidifying the molten metal as the molten metal flows onto the cooling surface, the method including capturing visual data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation, applying at least one machine learning model to the visual data to determine one or more operating parameters of the metal foil fabrication process.

2. The method of claim 1, further including pre-processing the visual data including any one or more of image segmentation, edge detection or boundary detection, feature extraction and object recognition.

3. The method of claim 1 or 2, wherein the one or more operating parameters of the metal foil fabrication process are determined in real-time or near real-time based on the visual data.

4. The method of any one of the preceding claims, wherein the visual data includes live video data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation.

5. The method of any one of the preceding claims, wherein at least one of the operating parameters includes a thermal distribution map associated with the metal foil fabrication process, and wherein the thermal distribution map includes a thermal distribution of the melt distribution receptacle, molten metal and cooling surface during operation of the planar flow caster.

6. The method of any one of claims 5, and the method further includespre-processing the visual data including clustering pixels based on mean cluster centroids to generate pixel clusters, and wherein the step of applying at least one machine learning model to the visual data to determine one or more operating parameters of the metal foil fabrication process includes determining one or more temperature values corresponding to each pixel cluster using a supervised machine learning model, the method further including generating the thermal distribution map based on the temperature values and pixel clusters.

7. The method of any one of the preceding claims, wherein at least one of the operating parameters is a variable distance between the melt distribution receptacle and the cooling surface.

8. The method of claim 7, further comprising pre-processing the visual data including determining a first boundary corresponding to the melt distribution receptacle, and determining a second boundary corresponding to the cooling surface.

9. The method of claim 8, wherein determining the second boundary includes determining the second boundary based on a reflection of melt distribution receptacle on the cooling surface, and optionally, the first boundary and the second boundary are determined using one or more unsupervised machine learning models.

10. The method of any one of claims 7 to 9, wherein the variable distance between the melt distribution receptacle and the cooling surface is determined based on a distance between the first boundary and the second boundary, and optionally the distance between the first boundary and the second boundary is determined using a supervised machine learning model.

11. The method of any one of the preceding claims, wherein at least one of the operating parameters is a flow velocity of the molten metal from the melt distribution receptacle towards the cooling surface.

12. The method of claim 11 further comprising pre-processing the visual data includingdetermining a region of interest between the melt distribution receptacle and the cooling surface, wherein a boundary of the cooling surface is determined based on a reflection of the melt distribution receptacle on the cooling surface.

13. The method of claim 12, wherein determining the flow velocity of the molten metal includes determining an intensity or colour threshold for image pixels representative of the molten metal, estimating a width of the pixels at or above the threshold within the region of interest, and estimating a height of the pixels at or above the threshold within the region of interest.

14. The method of claim 13, wherein determining the flow velocity of the molten metal includes calculating the flow velocity based on the estimated width and height.

15. The method of any one of the preceding claims, wherein at least one of the operating parameters is a flow rate of the molten metal from the melt distribution receptacle towards the cooling surface.

16. The method of claim 15 further comprising comparing frames of visual data and determine changes in pixel intensity values, identifying regions of interest indicative of flow of molten metal, and determining the flow rate based on area measurements of the identified regions of interest.

17. The method of any one of the preceding claims, wherein at least one of the operating parameters is speed of the casting wheel providing the cooling surface.

18. The method of claim 17, further comprising determine circular features of the casting wheel based on edge detection, generate velocity vectors based on the circular features, determine speed of the casting wheel based on the velocity vectors.

19. The method of any one of the preceding claims, further comprising determining one or more quality parameters of the metal foil.

20. The method of claim 19, wherein the quality parameters include one or more dimensions of the metal foil.

21. The method of any one of the preceding claims, wherein the metal foil is a shape memory alloy (SMA).

22. The method of any one of claims 1 to 20, wherein the metal foil is an amorphous alloy foil.

23. A computer-implemented method for determining control parameters for controlling operations of a planar flow caster in a metal foil fabrication process, wherein the planar flow caster includes a heating coil for heating a mass of metal during a metal melting process, and a casting wheel, the casting wheel providing the cooling surface, the method including determining one or more operating parameters according to the method of any one of claims 1 to 22, and determining one or more control parameters based on the one or more determined operating parameters to control operations of the planar flow caster, wherein the control parameters include any one or more of a temperature of the heating coil, a rotating speed of the casting wheel, a descending speed of the melt distribution receptacle, and a pressure generated during the metal melting process.

24. A planar flow caster, including a heating coil for heating a mass of metal during a melting process to create a molten metal, a melt distribution receptacle through which the molten metal is extruded, anda casting wheel, the casting wheel having a cooling surface for cooling and solidifying the molten metal as it flows onto the cooling surface from the melt distribution receptacle, and a camera operatively configured to capture visual data of the melt distribution receptacle, molten metal and cooling surface during operation.

25. The planar flow caster of claim 24, further including a process monitoring module operatively configured to process the visual data via any one or more of image segmentation, edge detection or boundary detection, feature extraction and object recognition, and apply at least one machine learning models to the visual data to determine one or more operating parameters of the planar flow caster in real-time or near realtime, the one or more operating parameters including any one or more of a thermal distribution map of the metal foil fabrication process including a thermal distribution of the melt distribution receptacle, molten metal and cooling surface during operation of the planar flow caster, a variable distance between the melt distribution receptacle and the cooling surface, a flow velocity or flow rate of the molten metal from the melt distribution receptacle towards the cooling surface, and speed of a casting wheel providing the cooling surface.

26. The planar flow caster of claim 24 or 25, wherein the visual data includes live video data of the melt distribution receptacle, molten metal and cooling surface of the planar flow caster during operation.

27. The planar flow caster of any one of claims 24 to 26, wherein process monitoring module is operatively configured to carry out a method according to any one of claims 5 to 22.

28. The planar flow caster of any one of claims 24 to 27, further including a closed-loop control module, the closed-loop control module being operatively configured to determine one or more control parameters to control operations of the planar flow caster based on oneor more operating parameters determined by the process monitoring module, wherein the control parameters include any one or more of a temperature of the heating coil, a rotating speed of the casting wheel, a descending speed of the melt distribution receptacle, and a pressure generated during the metal melting process.

29. A metal foil fabricated using the planar flow caster of any one of claims 24 to 28.

Images