Patent Application
20250135527
The present invention relates to a system, apparatus, and method for continuous casting of metal, and more particularly, to a system, apparatus, and method for the determination of parameter values for continuous casting of metal using machine learning and artificial intelligence to reduce or eliminate defects in the casting.
Metal products may be formed in a variety of ways; however numerous forming methods first require an ingot, billet, or other cast part that can serve as the raw material from which a metal end product can be manufactured, such as through rolling or machining, for example. One method of manufacturing an ingot or billet is through a continuous casting process known as direct chill casting, whereby a vertically oriented mold cavity is situated above a platform that translates vertically down a casting pit. A starting block may be situated on the platform and form a bottom of the mold cavity, at least initially, to begin the casting process. Molten metal is poured into the mold cavity whereupon the molten metal cools, typically using a cooling fluid. The platform with the starting block thereon may descend into the casting pit at a predefined speed to allow the metal exiting the mold cavity and descending with the starting block to solidify. The platform continues to be lowered as more molten metal enters the mold cavity, and solid metal exits the mold cavity. This continuous casting process allows metal ingots and billets to be formed according to the profile of the mold cavity and having a length limited only by the casting pit depth and the hydraulically actuated platform moving therein.
The continuous casting process involves hundreds of variables that impact quality of the resultant casting. Variables such as metal temperature, metal chemistry, cooling water temperature, flow rates, casting speed, and water chemistry, among numerous other variables can dramatically affect the quality of a casting. Optimizing parameter values for continuous casting is necessary to reduce and potentially eliminate defects in the casting. Such parameter value determination is challenging.
The present invention relates to a system, apparatus, and method for continuous casting of metal, and more particularly, to a system, apparatus, and method for determining parameter values for continuous casting of metal using artificial intelligence to reduce or eliminate defects in the casting. Embodiments provide a method of parameter value determination for direct chill casting including: obtaining at least one image of a surface of a casting that has been cast with a first set of casting parameter values and set-up conditions, where the at least one image of the surface of the casting includes at least one casting anomaly; processing the at least one image of the surface of the casting through a first machine learning model as an input; receiving, from the first machine learning model and a classifier, at least one classification as an output, where the at least one classification comprises at least one casting anomaly classification; Alternatively, directly inputting a casting anomaly classification or classifications and processing the at least one classification as an input to a second machine learning model; and receiving, from the second machine learning model, an indication of at least one of casting parameter value changes or set-up condition changes to be made to the first set of casting parameter values or set-up conditions to arrive at a second set of casting parameter values and set-up conditions, where the second set of casting parameter values and set-up conditions are intended to reduce or eliminate the at least one casting anomaly.
According to some embodiments, the first machine learning model is a deep neural network. The first machine learning model of an example embodiment is a classification engine. According to some embodiments, the at least one casting anomaly comprises one or more of: butt curl, cold folding, cracking, oxide patches, tears, folds, lap lines, liquation, surface pimples, surface blisters, profile, steam stains, spiraling, or bleed out/over. The first set of casting parameter values of some embodiments include values for one or more of: casting material temperature, casting material chemistry, water temperature, water chemistry, start water flow rate, water ramp rate, run water flow rate, casting start speed, speed delay, speed ramp and run speed, processing equipment preheat temperatures, metal level, metal level ramping, fill rate, pin position, or casting gas flow rate.
According to some embodiments, the method includes receiving, from the first machine learning model, at least one classification as an output where the at least one classification includes at least one casting anomaly classification and at least one corresponding size measurement and cast length location of the at least one surface anomaly. According to certain embodiments, the input further includes an indication of a size of the at least one casting anomaly.
Embodiments provided herein include a method of parameter value determination for direct chill casting including: receiving data corresponding to a casting defect of a casting that has been cast with a first set of casting parameter values; determining, from the data, at least one classification as an output, where the at least one classification includes at least one casting defect classification; processing the at least one classification as an input to a second machine learning model; and receiving, from the second machine learning model, an indication of casting parameter value changes to be made to the first set of casting parameter values to arrive at a second set of casting parameter values, where the second set of casting parameter values are intended to reduce or eliminate the at least one casting anomaly. According to some embodiments, determining, from the data, the at least one classification as the output includes determining, using a first machine learning model, the at least one classification as the output.
According to some embodiments, processing the at least one classification as the input to the machine learning model further includes processing the at least one classification along with the first set of casting parameter values as the input to the second machine learning model. The data of some embodiments includes at least one of image data or non-destructive analysis data. According to certain embodiments, the non-destructive analysis data is collected using one or more of photographs, laser profile testing, photogrammetry, linear displacement testing, Eddy Current testing, Magnetic Testing, Thermographic Testing, Resonant Testing, Radiographic Testing, or Ultrasonic Testing. According to certain embodiments, the at least one classification further includes a priority of the at least one casting defect and a severity of the at least one casting defect. According to certain embodiments, the first machine learning model is a classification engine.
Embodiments provided herein include an apparatus including at least one processor and at least one memory including computer program code, the at least one memory and the computer program code configured to, with the processor, cause the apparatus to at least: receive data corresponding to a casting defect of a casting that has been cast with a first set of casting parameter values; process the data as input to a first machine learning model; receive, from the first machine learning model, at least one classification as an output, where the at least one classification includes at least one casting defect classification; process the at least one classification as an input to a second machine learning model; and receive, from the second machine learning model, an indication of casting parameter value changes to be made to the first set of casting parameter values to arrive at a second set of casting parameter values, where the second set of casting parameter values are intended to reduce or eliminate the at least one casting defect.
According to some embodiments, causing the apparatus to process the at least one classification as the input to the second machine learning model further includes causing the apparatus to process the at least one classification along with the first set of casting parameter values as the input to the second machine learning model. According to some embodiments, the at least one classification includes a priority and a severity of the at least one defect. The priority and the severity of the at least one defect include measures determined based, at least in part, on a position and a size of the at least one defect.
Having thus described the invention in general terms, reference will now be made to the accompanying drawings, which are not necessarily drawn to scale, and wherein:
Exemplary embodiments of the present invention now will be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all embodiments of the invention are shown. Indeed, the invention may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. Like numbers refer to like elements throughout.
Embodiments of the present invention generally relate to the process of continuous casting of a direct chill casting mold to facilitate a more consistent ingot and/or billet. Vertical direct chill casting is a process used to produce ingots or billets that may have large cross sections for use in a variety of manufacturing applications. The process of vertical direct chill casting begins with a horizontal table containing one or more vertically oriented mold cavities disposed therein. Each of the mold cavities is initially closed at the bottom with a starting block or starting plug to seal the bottom of the mold cavity. Molten metal is introduced to each mold cavity through a metal distribution system to fill the mold cavities. As the molten metal proximate the bottom of the mold, adjacent to the starting block solidifies, the starting block is moved vertically downward along a linear path. The movement of the starting block may be caused by a hydraulically lowered platform to which the starting block is attached. The movement of the starting block vertically downward draws the solidified metal from the mold cavity while additional molten metal is introduced into the mold cavities. Once started, this process moves at a relatively steady state for a semi-continuous casting process that forms a metal ingot having a profile defined by the mold cavity, and a height defined by the depth to which the platform and starting block are moved.
During the casting process, the casting and the mold itself is cooled to encourage solidification of the metal prior to the metal exiting the mold cavity as the starting block is advanced downwardly, and a cooling fluid is introduced to the surface of the metal proximate the exit of the mold cavity as the metal is cast to draw heat from the cast metal ingot and to solidify the molten metal within the now-solidified shell of the ingot. As the starting block is advanced downward, the cooling fluid may be sprayed directly on the ingot to cool the surface and to draw heat from within the core of the ingot.
The direct chill casting process enables ingots to be cast of a wide variety of sizes and lengths, along with varying profile shapes. While circular billet and rectangular ingot are most common, other profile shapes are possible. Circular profile billets benefit from a uniform shape, where the distance from the external surface around the billet to the core is equivalent around the perimeter. However, rectangular or variable periphery ingots lack this uniformity of surface-to-core depth and thus have additional challenges to consider during the direct chill casting process.
Various complexities exist in the casting of metal parts, particularly in vertical direct chill continuous casting. There are well over 100 individual inputs or casting parameter values that impact the outcome of a continuous casting operation, with the potential for substantially more inputs particularly employing embodiments described herein which can better handle adjustment of inputs to improve the quality of a casting.
Currently, while casting (or shortly after the cast is finished), the cast billet or ingot is visually observed or inspected. If a defect is noticed, a manual process of defect identification and troubleshooting begins using information contained in defect guides that correlate defects with various causes. This process is laborious and arduous, particularly in view of how the various individual casting parameters impact the casting quality and the interaction between various parameters that can resolve one issue while causing another. Process improvement to overcome the defects being produced is generally a stepwise process. The most likely responsible casting parameter may be adjusted first. If that adjustment does not solve the problem, the next likely variable is manipulated, and so on. Because this process is done manually, it can take significant time to arrive at a solution. Further, it is generally not even possible to only change one variable at a time which means that a change to one casting parameter that affects another casting parameter renders the improvement or decline in quality of the casting difficult to pin down to a specific cause. Further, as it takes time to produce casting and record defect information, correlation back to a specific casting parameter can be challenging.
Embodiments described herein improve upon the manual process by implementing an automated parameter value determination process by using defect analysis and parameter value determination tools. The collected defect information, together with the casting parameters associated with the cast are analyzed to correlate specific defects with process and other influencing casting parameter values. As the database of information grows in the form of training data for a machine learning model, the machine learning model will learn and become more efficient and effective in predicting defects, allowing for process improvements much earlier in the process.
Embodiments of the present disclosure collect the inputs or casting parameter values for a casting operation and then apply multiple techniques to the input values to categorize, prioritize, and relate the individual input values to specific internal and external casting defects. These correlations are then analyzed by artificial intelligence (AI) through machine learning to determine improvements to the existing techniques, providing for adjustment to improve all quality measures of the cast product including process timing determination. The as-cast product is analyzed, attributes collected, and defects identified. Techniques then use the identified defect and ingot/billet attributes to classify the defect by name, priority, severity, and magnitude to provide recommended process, casting practice, chemical, and equipment control changes to any one or combination of the 150-plus identified casting parameter inputs in one example embodiment.
The machine learning models as described herein can be implemented on a computer implementing a processor to apply the machine learning models and techniques thereof. The machine learning models can access training data and learned data stored, for example, on a memory device. Embodiments described herein can thus be implemented on a non-transitory-readable medium storing computer with executable program code instructions to execute the machine learning models described herein.
A direct chill casting mold to produce an ingot with a rectangular profile does not have a perfectly rectangular mold cavity due to the deformation of the ingot as it cools after leaving the mold cavity. The portion of the ingot exiting the mold cavity as the platform and the starting block descend, retains a molten or at least partially molten core inside the solidified shell. As the core cools and solidifies, the external profile of the ingot changes such that the mold cavity profile, while it defines the shape of the final, cooled ingot, does not have a shape or profile that is identical to the final, cooled ingot.
Direct chill casting molds, such as molds produced by Wagstaff, Inc. of Spokane Valley, Washington, can include mold walls that have variable shape profiles. Such variable shape profiles enable the mold cavity to have a profile that changes based on a phase of a casting process, which can reduce waste by producing ingots that have a substantially uniform profile along their entire length. The mold sidewall shapes and dynamic changes thereto provide additional inputs that can be leveraged by techniques noted above in mitigating defects and improving the overall quality of a casting as described herein.
The various aspects of the direct chill casting process, such as the speed of descent of the starter block 215, the flow rate of molten metal through the spout 230, the temperature profile of the molten metal at various stages along the flow path into the casting, the temperature, chemistry, and flowrate of cooling fluid sprayed on an external surface of the casting, the temperature profile of the mold walls 205, the casting chemistry, and numerous other inputs all have an impact on quality of the cast product. Thus, all of these variable inputs can be tuned to produce the highest quality cast product.
Defects of a continuous casting cast product can be of a wide variety of types and severities. One example of a defect is a surface crack. The detectability of a surface crack is substantially greater than internal cracks. Internal cracking may be detected through a variety of testing processes, such as Eddy Current testing, Magnetic Testing, or Thermographic Testing for near-surface defects, or Resonant Testing, Radiographic Testing, Ultrasonic Testing, or the like for defects further from a surface of the casting. Other defects, such as butt swell, surface bubbling, surface ripples, tears, folds (vertical and cold), bleed-out/over, cold folding, lap lines, liquation, surface pimples, surface blisters, spiraling, distorted profile, cooling rings, water stains, metal tags, etc. can be identified often through surface analysis, though sometimes require microscopic analysis of the surface.
Currently, while casting or shortly after the cast is finished, the cast billet/ingot can be visually observed or inspected. This inspection process can include surface analysis using the naked eye, or some magnification apparatus to observe finer details of the surface. The inspection can also include an inspection of an internal cross-section of the casting, where the casting can be cut to expose parts of the casting not visible from the outside of a complete casting. Further, inspection can include the testing processes described above that can include non-destructive analysis of the internal structure of the casting. If a defect is noticed, a manual process of defect identification and troubleshooting begins. This can be a laborious and arduous process. Improvements in the process to overcome defects is generally a step wise process. The most likely variable may be adjusted first, and if the problem is not solved, the next likely variable is manipulated, and so on. As this process is done manually, it can take numerous iterations and substantial time to reach a solution. Further, it is generally not possible to only change one variable at a time, which means that if one parameter value is changed, some other parameter values are also changed, and it is unclear what actually solved the problem. Further, because of the time it takes to record defect information and then go back and try to correlate the defect with process variables that may have caused the problem, the time required to make this determination may be too long for it to be worth attempting.
Embodiments described herein employ machine learning and artificial intelligence to monitor for defects, analyze the defects, and determine solutions to mitigate the defects. Embodiments employ sensors and software used for defect analysis, then implement additional sensor technology for continuous process parameter measuring and recording. Together this data is analyzed using analytical techniques that correlate defects with process parameters and their respective values along with other influencing parameters. As the database of information concerning defects and resolutions grows, the machine learning model of example embodiments will learn and become more efficient and effective in predicting failures allowing for process improvements much sooner in the process and in a shorter amount of time, with the goal being to process corrections before a significant defect occurs.
Described herein are a number of defects for cast products and a number of inputs that can be used to adjust the casting operation. However, these defects and inputs are not exhaustive, and other defects and inputs can be used without deviating from the scope of this disclosure. By collecting and analyzing the data automatically, the process of collecting and analyzing data will become more efficient and effective. Employing software techniques described herein on the collected data enable a more though analysis showing the interdependencies and interactions of the input variables.
Identification of defects in a casting through analysis of collected data can be performed in a variety of ways. Image analysis for external surface defects can be performed to identify issues that are visible on the external surface of the casting. Image analysis can also be performed in an analysis of a cross-section of the casting; however, requiring the casting to be cut to obtain a visible cross-section surface which is not practical for normal production. Non-destructive analysis can be performed using Eddy Current testing, Magnetic Testing, or Thermographic Testing for near-surface defects, or Resonant Testing, Radiographic Testing, Ultrasonic Testing, or the like. These types of analysis are referred to as “non-destructive” as they do not require the casting to be cut, segmented, or otherwise compromised for purposes of the analysis.
In order to identify defects in a casting through image and non-destructive testing analysis, defects need to be identified in a set of training data. This training data can include identification of a defect that is confirmed through a manual review process. Images and non-destructive test data can be analyzed, via manual user review or by image/data analysis tools. In manual review, a human operator can identify defects and classify the defects according to a defect or an anomaly classification. For image/data analysis tools, defects can be identified within the images or data, and presented to a human operator for review. The human operator can confirm or reject the identified defect. Further, the image/data analysis tool can attempt to classify the defect with an anomaly classification. This may be performed by a machine learning model; however, in order to establish sufficient training data, this training data may be a supervised machine learning model, whereby the identification and classification of defects is reviewed and confirmed/rejected by a human operator. This process can establish a corpus of training data from which defects can be properly identified and classified without requiring a human operator. Further, as a machine learning model receives input data for processing by the machine learning model, this data can be added to the training data corpus to better refine the data and image analysis process and improve the accuracy and repeatability of the defect identification and classification.
Upon identification and classification of defects through artificial intelligence employing the trained defect detection machine learning model, the defect can be used as an input to a second machine learning model, whereby the second machine learning model uses the defect to determine how to adjust casting operating parameter values to mitigate or eliminate the defect.
The second machine learning model requires initial training to correlate certain operational parameters with specific defects. This training can occur using supervised machine learning where techniques are initially identified that correlate defects with operating parameter value adjustments to remedy the defects. A human operator can confirm output of the supervised machine learning model to build a corpus of training data that is known to be valid.
According to an example embodiment, a casting pit operational and control architecture can employ at least twelve key variables: speed practice (start speed, start speed delay, speed ramp rate, stop speed), water practice (start water, start water delay, water ramp rate, run water flow rate), metal level practice (start metal level, steady state metal level, and all ramps between metal levels), fill rate (for each position), pin opening (for each position), cast water temperature, cast water properties (alkalinity, turbidity, conductivity, pH, etc.) casting launder hot and cold end temperature among other potential parameters.
Speed practice relates to the speed of a starting head descending into a casting pit. The casting head movement within a casting pit generally has a speed profile, whereby the starting head starts in a static position, and ramps up to a starting speed for a predetermined time or distance of travel. The starting head achieves a steady state speed during the casting operation, and then ramps down in speed at an end of a casting to again achieve a static position when casting is complete. The start speed, start speed delay, speed ramp rates, run speed, etc. all contribute to properties of the cast ingot or billet.
The water practice relates to the parameters surrounding the introduction of water through the mold to the casting. The water practice can include start water, where the starting time, temperature, flow rate can be identified. The start water delay can include a delay or maintaining of start water flow rate until water ramping begins. The water ramp rate can include a ramp rate of flow rate of the water at the casting and at an end of the casting operation. The run water can include a flow rate of the water during steady state operation of the casting process. Water temperature is critical throughout the cast. Some other water parameters that impact the casting operation include the cast water alkalinity, cast water turbidity, cast water conductivity, and the cast water pH as well as others.
The metal level practice relates to the molten metal level within the continuous casting mold. The metal level is generally controlled to a specific value but can be affected by speed of descent of the starting head, defects such as flashing, bleed-outs, bleed-overs, or hang-ups all of which impact the flow rate of metal into the continuous casting mold. In modern casthouses, there is typically a metal level practice with a starting metal level at the beginning of the casting operation, a steady state metal level during the casting operation, and ramps between the different metal levels within the continuous casting mold. The fill rate of the continuous casting mold determines the metal level of the mold prior to the platform 220 starting down. Total metal flow rate after start down is determined based upon the speed of descent of the starting head and the metal level to be achieved. The pin opening is the pin within the spout, whereby the pin opening affects the flow rate of metal through the spout into the continuous casting mold. The cast launder hot and cold temperatures relate to the launders that carry molten metal ahead of reaching the continuous casting mold. These can be heated to achieve a temperature that helps maintain the molten metal at a specific temperature ahead of casting.
In addition to the above-identified parameters, numerous other variables impact the outcome of a casting operation. According to embodiments described herein, techniques of the machine learning model classify each defect according to priority, severity, and magnitude to provide recommended process, casting practice, chemical, and equipment control changes to any one or a combination of the parameters and parameter values of the casting operation.
The machine learning models provided herein include the defect identification/classification model, and the parameter value determination model. The defect identification/classification model uses as input, images of the casting surface and data from the non-destructive testing methods to identify defects and classify them according to their defect type, priority, severity, and magnitude as the output of the model. The parameter value determination model uses as inputs the output of the identification/classification model for a casting operation. The parameter value determination model, based on the defect type, priority, severity, and magnitude along with the casting parameters associated with the ingot or billet from which the defects were identified outputs recommended process, casting practice, chemical, and equipment control changes for the casting operation. Between these two models, the casting operation is improved upon. The improvements may, in some embodiments, be iterative, particularly as the parameter value determination model learns from the improvements to the casting parameter values.
Casting parameters and set-up conditions, as described herein, can include various aspects of the casting process that can be adjusted based on a variety of conditions. Casting parameters and set-up conditions can include items in the tables of
The defect detection/classification model can collect attributes relating to the casting to inform the techniques used to classify the defects. These attributes can include an abort length at which point the casting was aborted, a casting length of the overall casting, a butt crop length of a distance from the start of the casting for cropping of the butt, scalp depth from the top of the casting, grain size of the metal grains, secondary dendrite arm spacing, shell zone, macrosegregation levels, surface micro-segregation levels, steamlines and shapes, steamstains and shapes, as well as corner butt curl. Secondary dendrite arm spacing (SDAS) is a parameter for the determination of local quench rate of a casting. The shell zone is a relatively thin band of elemental segregation near the surface of the casting caused by differing solidification rates. Macrosegregation is the elemental segregation from surface to center of a casting. Micro-segregation occurs when solute concentration is not constant through the casting.
Defects in the casting can include defects such as large grains, large micro-segregation above a predetermined limit, large dendrite arm spacing, macrosegregation above a predetermined limit, steamlines above or below a predetermined limit, corner butt curl above or below a predefined limit, any cracks, bleed-out, bleed-over, flash, tear, drag, exudate, cold fold, edge, or corner lift, for example.
The geometrical features of a casting can be captured and analyzed relative to an ideal standard, such that the geometrical features can be deemed as defects if not maintained at the ideal standard or within a predetermined degree of accuracy. The parameter value determination model may use, as input, the geometrical features to identify parameter values for determination. Geometrical features can include thickness, width, length, profile, twist, lateral bow, longitudinal bow, butt curl, and butt and head shape scans for a cast ingot or billet. This information can be achieved through vision or laser measuring and cataloging to capture data and provide geometrical feature data back to the parameter value determination model to identify how certain parameter values impacted the geometrical features of a casting.
According to embodiments described herein, the defect identification/classification model can include various modules within the model to address specific types of defects. For example, a crack may be readily discernable from the image analysis or data analysis from the non-destructive testing. The cracking technique module can sort and grade cracks. The position of the cracks on the casting is also informative of the defect type and how it may be remedied in the parameter value determination model.
Failure modes beyond cracking can occur during continuous casting, some of which are more readily identified than others. Described herein are defect types that are generally applicable to ingots, or castings having a generally rectangular profile. Some of these defect types apply to billets having a generally circular profile; however, billet defect types will be addressed further below. For ingots, a bleed out failure or defect is any loss of molten metal through a gap or tear in the ingot shell. A melting or tearing of the shell below the meniscus that dumps molten metal out of the ingot sump is considered a bleed out. Bleed outs can be caused through a variety of issues including a damaged or rough casting bore, excessive speed ramping of the starting head, shallow water ramping of the cooling water, excessive metal head height, or poor metal distribution. Each of these defects can be identified and classified using a first machine learning model. The first machine learning model receives, as input, images or data from non-destructive testing associated with a casting. A defect or anomaly is identified in the images or data using the first machine learning model.
The first machine learning model is trained on a corpus of training data including images and data from non-destructive testing with positively identified defects or anomalies along with their respective types or classifications. Once trained, the first machine learning model is then able to identify a defect or anomaly in the images or data, and to classify the defect or anomaly according to a type or classification of defect or anomaly. The first machine learning model can further identify a severity of the defect.
A defect type of a bleed over on ends is an escape of metal over the top of solidifying metal caused by deformation of the ingot butt (e.g., butt curl) or rapid solidification which occurs when water contacts the ingot as soon as it emerges from the mold. Bleed over on ends can be caused by, among other factors, a cold condition, broken metal distribution bag or poor metal distribution, cold metal, inadequate metal level during secondary butt curl, shallow ramp speed, or excessive water ramping.
Horizontal cracks or tears are generally caused by butt bounce of the casting. The butt bounce is generally caused by the vaporization of water between the bottom of the ingot or billet butt and the starting head. Horizontal cracks or tears can be caused by a lack of starting head water drainage, plugged drain holes, excessively hot start practice, rapid water ramps, or other contributing factors.
Another type of defect is a cold fold, which is a condition where the advanced cooling distance repeatedly freezes the meniscus in the mold, but not severely enough to cause a bleed over. New molten metal fills in behind and continues in a repetitive manner. Cold folds can occur as the result of a cold condition, broken metal distribution bag, poor metal distribution, inadequate metal level, shallow speed ramp, excessive water ramp, or other contributing factors.
A collapsed butt defect is a fracture that starts in the dish out/rim area as a result of insufficient strength to support the ends curling off the block. These cracks are identified by the elevated rim where the crack starts. Collapsed butt defects can occur from, among other factors, inadequate film boiling length (e.g., the steam barrier is not long enough) or the fill time is not long enough.
A corner dribble defect is a condition caused by metal freezing in the corners and new metal bleeding over the top of the solidified metal. Causes of corner dribbles can include rapid secondary curl, cold butt, too much cooling in the corners, a cold condition, broken metal distribution bag or poor metal distribution, inadequate metal level, shallow speed ramp, or excessive water ramp among other factors.
Drag type defects occur when molten aluminum sticks to the mold bore. Drag defects have sufficient surface texture to be visible even over exudated surfaces. Such a defect runs parallel to the cast direction and is generally one to ten millimeters wide. Drags can occur from a variety of factors including: burnt oil or residue on the graphite bore, use of the wrong type of oil, oil that is not adequately absorbed into the graphite liner, a defect in the graphite bore, or a metal tag or damage on the aluminum bore.
Defects in the form of edge lift defects occur at the start of the cast when the periphery of the ingot butt adheres to the mold bore. Edge lift creates a hot spot that can cause cracking or even tear the ingot shell, causing a bleed out. Edge lift can occur from, among other conditions, a starting head not engaged into the mold at the proper position at the start of casting, the mold not being level, starting head engagement being uneven, or the mold bore being damaged.
Flashing type defects can occur along the edges and rolling faces of an ingot, resulting from molten metal entering and solidifying in the gap between the mold and starting head. This can occur when the gap is excessive. Flashing type defects can occur from, among other factors, a mold not properly being aligned with the starting head, the mold not being level, the starting head not being level, or a starting head that has distorted in width over extended use.
A hangup type defect is a condition in which the ingot stops in the mold while the platen and starting head continue down. This is a dangerous situation that can lead to ingot kinks and can potentially submerge molten aluminum in water when the ingot releases from the mold. Hangup type defects can occur for a variety of reasons including: excessive curl, excessively hot conditions, bleed over, negative mold bore taper, or mechanical damage to the mold bore.
A hot butt separation type defect is caused by the cast being started before the butt is sufficiently solidified and strong enough to withstand the starting forces. This can occur when the hold time is too short before starting the starting head movement, the metal is too hot, or from a heat resistant coating on the starting head during a hold/fill sequence.
A hot crack at transition type defect is a separation of solidifying material initiated in the mushy zone (metal coherency zone) that occurs after the platen or starting head starts advancing down, and usually begins during the transition stage where the ingot transitions from start to run parameters. This type of defect often occurs during ramping of speed and water. A hot crack at transition defect can occur when ramping open of the water valve is too late or the total flow of water during the transition is too low. This type of defect can also occur when the water temperature is too high for the water ramp in the current casting process, or the casting speed ramp up is too fast.
A hot crack type defect is a separation of solid material initiated in the mushy zone. This type of crack is generally from an exudate pocket, fold, bump, or smear that forms a stress riser on the thin, hot shell. Heat input may be decreased or cooling increased at the start to avoid such a defect. Some causes of a hot crack type defect can include starting water flow is too low, starting speed or the speed ramp rate is too fast, start metal temperature is too high, water quenchability has changed, or the mold water pattern is defective.
A hot tear type defect is a melting or tearing of the shell below the meniscus. This can be caused by a damaged or rough casting mold bore, excessive speed ramp, shallow water ramp rate, insufficient lubrication, excessive metal head height, or poor metal distribution, for example.
A nuisance crack is a type of defect that includes short cracks near the corner caused by a dissimilar cooling rate between ends (short ends) and the rolling face (long sides). This typically occurs when water quench begins. They can generally be mitigated by allowing more time for the butt to solidify in the mold against the starting head prior to contacting the water.
An oxide patch crack type defect is a flake of aluminum oxide. Localized pockets of extra aluminum oxide float on the head of the ingot until pulled over the meniscus by the movement of the oxide layer on the molten metal. Oxides are insulators and they deflect water, causing a hot spot and stress riser on the shell where cracks may initiate. These oxide patch crack type defects can be caused by moisture or contamination in the starting head creating an oxide patch that is not constrained by the skim dam. Causes can also include cracked, wet, or broken skim dams, or skim dams that are not level or do not penetrate deep enough into the metal head. Causes can further include improper metal head skimming methods.
A water stain type defect can occur from mineral deposits left on an ingot due to film boiling. Water staining is a precursor to a cracked ingot and should be addressed quickly. Causes can include damaged water spray holes in the mold, or an inconsistent or weak water pattern from plugged or shrunk spray or baffle holes.
Kinked ingots or bent ingots are defect types caused by either thermal imbalances or mechanical problems. Causes can include flashing or bleed outs that held the ingot to the starting head at the cast start. Causes can further include lateral shifts in the starting head during descent or uneven but curl. Causes can still further include ingot hang up in the mold due to excessively hot conditions, inadequate mold bore taper, or inadequate mold bore lubrication.
Defect types can include wavy ingots which is a repetitive surface variation across the rolling face of an ingot with a definitive periodicity and amplitude to the waves. This can be caused by excessive oil, excessive run speed, insufficient water flow, or excessive metal levels, for example.
Butt cold cracks are post solidification type defects that start in the butt and typically exhibit on both rolling faces due to the brittle nature of aerospace alloys and extreme contraction forces found in the butt. A massive separation of the metal can occur at any time after solidification. Causes can include that the butt is too weak or has cooled too quickly. The curl transition can be too harsh or the metal chemistry or grain refinement could be poor. In some cases, the alloy is too brittle or the water too cold. Too much water under the ingot at cast start can cause cold cracks. Other causes can include a wiper that is too low, insufficient butt pad size, start defects, or stress risers on the butt such as from drain plugs.
While the above described defect types generally relate to ingot defects, defects of both similar and distinct types occur in billets having a substantially round cross section. While the continuous casting process for billets is similar to that of ingots, the differing cross section and often different sizes lead to distinctions in the process. Thus, some defects, while similar to those of ingots, may have different potential causes. Billets are often cast on mold tables that support a matrix of rows and columns of billet molds. In such cases, the matrix of billet molds may receive molten metal at different flow rates, temperatures, or other disparities that can lead to different problems among the different molds.
Flashing type defects in billets are generally caused by molten metal penetrating the gap between the mold bore and the starting head during the fill operation. This condition is most apparent on billets nearest the metal entry because of higher metal temperatures and velocities. Potential causes of flashing type defects in billets include: molten aluminum entering the mold cavity with too much velocity or turbulence, metal temperature is too hot during the fill, the mold bore is distorted causing a gap with the starting head, the starting head is out of level with the mold bore, the starting head base or mold table is not level, the starting head is not properly engaged or centered with the mold, or the starting head is not within specification. Other causes may also lead to flashing type defects in billets.
Billet butt freeze-in type defects occur when aluminum or other metal freezes completely inside the mold and thimble opening, and the pull of the starting head is insufficient to dislodge the butt from the mold. This may occur on mold positions that are first to fill and have the longest hold times. Among the potential causes are the fill and hold time is too long causing metal to freeze in the mold and thimble opening, cold metal temperature during the fill, or mold damage, for example. Mold damage such as a burr or gap between the transition plate and casting ring or the casing ring and the mold bore may lead to butt freeze-in.
Cold butt separation type defects in billets may occur during the fill and hold operation. This may be caused by metal solidifying inside the thimble opening. In this case, at cast start, molten aluminum does not fully enter into the mold cavity until the solid butt clears the thimble opening. The molten aluminum then floods into the mold cavity and pours over the solid butt, creating a separation. The flooding aluminum can cause flashing that hangs up on the mold bore and a large gap occurs between the solid butt and the billet. Sometimes molten metal bleeds out under the mold bore. This appears similar to metal flashing; however, the bleed out occurs along the butt separation. Causes can include, among others, that the fill and hold time is too long, causing metal to freeze in the mold and thimble opening, cold metal temperature during the fill, or the trough not being sufficiently preheated.
A hot butt separation type defect can be caused by the cast being started before the butt is sufficiently solidified and strong enough to withstand the starting forces. Separation occurs at the top lip inside of the starting head. Molten aluminum can bleed out from this separated area, and looks similar to metal flashing. This type of defect can be caused by: hold time being too short (typically on the last molds to fill), the metal is too hot, mechanical damage to the mold bore, start speed being too fast, casting rings are damaged, or there is build-up of residue in the cavity of the starting head.
Cold folding type defects or cold shuts can occur when the metal meniscus solidifies near the transition plate due to cold conditions. As the solid metal moves away from the transition plate, molten aluminum or other metal floods into the cavity. The process repeats until the temperature condition in the mold becomes hotter and the solidification front moves away from the transition plate, allowing steady state casting to begin. Possible causes include: cold casting conditions, cold water temperature, irregular gas flow through the casting ring, poor transition plate/casting ring joint allowing casting gas to escape, or due to graphite coating the casting ring.
A bleed out type defect is a condition where molten aluminum runs out of an opening in the surface of the billet and spills into the casting pit. Signs of a bleed out include a hissing sound under the mold table and a drop in metal level in the trough. A vortex above the mold indicates that particular mold needs to be plugged off. Bleed outs can occur at any time, but most often occur at the start of a cast. For bleed outs occurring at the start of casting, possible causes include: that the hold time is too short resulting in the butt not being solid or strong enough at start; mechanical damage to the mold bore; butt freeze that can result in a bleed out if the butt falls out or remelts; a sudden drop of the platen at the start of the casting; or metal freezes in the gas pocket or hangs up in the casting ring/transition plate. Bleed outs occurring after cast start can be caused by: a butt freeze-in drops sometime during the cast; insufficient cooling of the billet surface, generally associated with water staining or heavy vertical drags; a mechanical condition inside the mold cavity that is tearing the billet surface; or cast speed is too fast.
Defects that include butt cracking occur when internal stresses exceed the strength or ductility of the material. Stress can be generated during the cooling and shrinkage of a solidifying butt. The stresses are often most severe during the start of a cast when the butt is solidified by heat extraction through the starting head and water quench. Crack prone alloys require a more conservative start practice. Some causes of butt cracking with internal crack type defects include an insufficient grain refiner, hot casting conditions, crack sensitive alloys, uneven fill procedures, uneven water cooling, or build-up of residue in the cavity of the starting head. Butt cracking with external crack type defects can be caused by hot crack prone alloys, insufficient water at start, or a hot butt.
Oxide patch type defects can be caused by oxides that accumulate inside the mold and then release and deposit on the billet surface. An oxide patch is a surface defect and typically does not extend below the billet surface. An oxide release at only the butt can occur as the main source is molten aluminum cascading into the mold during fill. Agitation of the molten aluminum by excess casting gas promotes the release of oxides during the start. Generally, larger molds produce larger oxide releases. Magnesium containing alloys can produce thicker oxides. Oxide releases randomly released along the full length of the billet may be due to a problem with the metal quality, excessive cast gas, moisture, or insufficient oxygen in the casting gas. Oxide patches only at the butt can be caused by: moisture in the trough, thimble or transition plate; bubbling from the mold or a rocking in the trough promoting oxide release; or poor surface conditions of the transition plate, such as roughness, coating, or joints that retain oxide in the mold. Oxide patches along the full length of the billet can be caused by: insufficient oxygen in the casting gas, melt preparation and fluxing practices generating salts carrying oxides, excessive bubbling from the mold due to high gas flows, moisture in the casting gas, or moisture in the transition plate.
Defects can include a situation where an air gap between the walls of the mold and the casting is compromised. This can occur when the gas bearing between the aluminum and casting ring is lost and the aluminum contacts the casting ring causing a deterioration of the billet surface quality. Friction between the billet and the casting ring can eventually lead to a tearing of the billet surface. The longer a billet is in contact with a mold wall, the more oil is consumed making it easier for the molten aluminum to attach to the casting ring. Common causes can include: low casting gas flow; casting conditions too hot; casting conditions too cold; loss of casting gas from the joint between the mold bore and the casting ring; insufficient casting oil; excessive casting oil which can restrict gas flow; oxide buildup at the transition plate; burned oil on the graphite ring; loss of casting gas from the mold cavity; or leaking casting gas system.
Horizontal tear type defects are often caused by molten aluminum that attaches to the inside of the mold and can result in a bleed out. Common causes can include: release coating on the casting ring surface; metal attached to the casting ring/transition plate joint; low casting gas or oil flow in a section of the casting ring; mechanical damage to the casting ring; or damage or cracks in the transition plate.
A vertical tear or zipper type defect is a deep scratch in the billet surface, often caused by a protruding obstruction inside the mold that touches the semisolid billet. This can be caused by a burr on the aluminum mold bore that tears the billet surface as it leaves the mold. Common causes can also include a metal buildup attached to the casting ring and transition plate joint which alters the metal flow as it solidifies, metal attached to a cracked or damaged casting ring surface, or metal in the gap between the bottom of the casting ring and the mold bore.
Lap lines are a type of defect that includes small, closely spaced horizontal lines that may be pronounced enough to be felt. Common causes can include a high casting gas flow causing rocking or bubbling in the trough or a rough or jerky platen. Liquation type defects occur when the billet surface has reheated inside the mold which causes the area to remelt, forming a rough texture or welts on the cast surface. Liquation occurs on alloys with a wide freezing range and can be caused by improper thermal balance.
Speed crack type defects can occur when the casting speed is high enough to cause a large temperature difference between the center and the outer surface of the billet. Non-destructive testing can be used to determine where the crack starts and ends. Common causes can include: excessive start or run speed, plugged water filter screens, plugged holes in the mold, ineffective grain refiner, hot casting conditions, or crack sensitive alloys.
Surface pimple type defects are often caused by high hydrogen levels in the molten aluminum. Hydrogen bubbles penetrate the surface of the billet during solidification and leave a rough sandpaper like surface. Pimple concentrations that are high at the start and diminish as the cast progresses may be due to moisture coming from the trough. Surface pimple type defects proximate the butt can be caused by moisture in the trough or inadequate degassing for the initial metal flow rate, for example. Surface pimple type defects along the length of the casting can be a result of inadequate metal degassing or inadequate water quenching, for example.
Surface blister type defects may appear on the surface of the billet after homogenization and may be formed by small pockets of expanding hydrogen. Blister concentrations that are high at the butt of the billet and diminish along the length may be due to moisture coming from the trough. Surface blisters proximate the butt may be caused by moisture in the trough or inadequate degassing for the initial metal flow rate, for example. Surface blisters along the length of the billet may be due to inadequate metal degassing or inadequate water quench, for example.
Spiraling type defects or “barberpoling” type defects are a winding depression on the billet surface that can resemble a screw thread. This occurs mostly on smaller diameter billets during the last 50% of the cast length. Often this is caused by a mechanically unstable platen or hot cast conditions. Spiraling in a single mold can be caused by plugged water passages or mold bore distortion affecting the water pattern. Spiraling where the defect stays with the table can be caused by plugged water passages or an unstable starting head. Spiraling on many molds can be caused by high water temperature, high metal temperature, hot casting conditions, or an unstable or jerky platen.
Distorted cooling ring type defects are noncircular bands on the head of the billet showing variations in billet solidification. This is not a defect by itself but indicates a water cooling problem. Possible causes include a weak or broken water pattern caused by plugged waters screens or mold water passages or by a distorted mold bore.
Water stain type defects are a surface condition often caused by calcium carbonate depositing onto the billet surface. It is usually a result of a hot billet surface condition and is amplified by hard water. Extreme cases can result in localized defects such as drags, welts, or small horizontal tears on the billet surface. Water staining occurring on a single mold can be caused by broken mold water patterns caused by blocked water passages or weak mold water patterns due to plugged water filter screens. If the staining occurs on many molds, the causes can include plugged water filter screens, low water flow rates, or high water temperatures.
Vertical drag type defects can be caused by metal sticking to the casting ring surface and disrupting smooth formation of the billet surface. Alloys of 2XXX and 7XXX series aluminum are very prone to dragging. If drags become severe due to hot casting conditions, a bleed out can occur. Causes of vertical drags can include: cold casting conditions, hot casting conditions, metal attachment in the casting ring/transition plate joint, graphite coating left on the casting surface causing metal attachment, gas flow too low, oil flow too low, low oil flowing through the casting ring, oil flow is too high with excess oil in the casting ring causing a reduction in gas flow, poor metal quality causing oxides to attach more readily to the mold interior, broken or weak water pattern, mechanical damage to the casting ring, or insufficient oxygen in the casting gas.
Bent billet defects have either a very gradual bow over the entire length of the billet or a sudden kink or bend. Bent billets are often caused by either a thermal imbalance or a mechanical problem respectively. Uneven water flow around a billet circumference can cause internal stresses, gradually bending the billet. Sudden bends in the billet are often due to mechanical disturbances, such as irregular platen travel. A thermally caused bent billet can be caused by a broken or weak water pattern causing uneven cooling or hot casting conditions. A mechanically bent billet can be the result of an unstable starting head causing bending to a specific mold position, bending at all mold positions due to a mechanical misalignment or shift in the platen guidance system, a gradual bend over the entire billet length due to platen and guide misalignment, a kinked billet indicating the butt lifted from the head during the start, then repositioned itself during the cast, mold table or starting head base out of level, or a grid guard is distorted and separates from the starting head base unevenly causing platen movement.
Metal tags on the billet head are another defect type that forms due to molten aluminum sticking to specific locations inside the mold. At the end of the cast, the stuck aluminum raises above the head until the sump solidifies. Metal tags disrupt the even flow of molten aluminum and create a cooler casting condition on the billet and hotter condition on the mold. These conditions create surface defects such as vertical disturbances. The location of the metal tag corresponds to the problem area in the mold that may require maintenance. Causes can include improper application of release coating on the transition plate, metal attaching to a chip or crack in the transition plate during the cast, or poor metal quality for example.
Oxide buildup on the billet head is another defect type that forms a black ring that accumulates inside the metal tag. The oxide buildup can break loose as a patch and deposit on the billet surface. In extreme cases, oxide buildup can prevent the billet from forming a proper air gap at the beginning of the cast or cause the air gap to be compromised.
As evident from the aforementioned defect types and potential causes, various defects can occur from a variety of factors. Further, there is substantial overlap in the contributing factors contributing to various different types of defects. Thus, it is challenging to discern what specific factors may contribute to a specific defect type. Embodiments described herein employ a machine learning model that uses techniques and training data to identify contributing factors while eliminating other potentially contributing factors to provide accurate parameter value determination for mitigating defects and improving the quality of castings. The machine learning model and techniques can consider the interplay among the various parameters when determining the cause of a defect type, and recommend parameter value or set-up changes for parameter determination.
The following figures relate to techniques employed in machine learning models to receive, as input, a defect type, location, and magnitude to establish how to mitigate the defects through parameter value determination.
To sort and grade cracks, the cracks may be given a name as in a type of crack along with a corresponding position, a priority (e.g., 1-25), a level of severity (e.g., A (least severe) to C (most severe)), and a quantity. For example, a “transition valve crack 5.C.3” is a crack with a level 5 priority (of 25), level C severity (severe), and a quantity of three.
Conversely, at the bottom of the listing of
Identifying the defects, such as defects listed in
According to an example embodiment, a technique of the parameter value determination module may receive as input the identification of a nuisance crack of less than two inches at the rim of a casting. The technique may first establish if curl is less than 30 millimeters. If the curl is less than 30 mm and there are not other defects, the casting may be considered relatively successful and no action is taken in the parameter value determination technique. If curl is greater than 30 mm and there are no other defects, the technique may then check grain refiner inoculation. If the inoculation to troughs was typical for the cast alloy and size combination, it may be established that inoculation was not an issue. If the inoculation to troughs was not typical, the parameter value determination technique may alert operators of improper trough inoculation. Similarly, several other parameters may be checked or verified to confirm if they were properly set or performed for the casting operation relative to prior successful casting operations.
For example, a grain refiner feed rate may be confirmed, a fill time of the mold may be confirmed, water temperature, metal temperature, starting head speed, etc. can be compared against those same parameter values that were used during successful casting operations. If any of the parameter values were not properly set or applied, the parameter value determination model can output corrective actions and parameter values that are believed to improve the quality of the casting through reducing the defects. Successful castings with little or no defects can be used as inputs to the second machine learning model, along with their associated casting parameter values. These are benchmarks of successful combinations of parameter values that led to a quality casting. The second machine learning model also uses as inputs identified and classified defects along with the parameter values associated with their castings. These inputs can teach the machine learning model what parameters can impact the quality of the casting to better inform how parameter values may be changed to minimize defects or anomalies.
A technique to identify operating parameter value adjustments and determination for an identified nuisance crack of less than two inches at the rim is illustrated in
An operation of an example apparatus will herein be described with reference to the flow chart of
In an example embodiment, an apparatus for performing the methods of
Many modifications and other embodiments of the inventions set forth herein will come to mind to one skilled in the art to which these inventions pertain, having the benefit of the teachings presented in the foregoing descriptions and the associated drawings. Therefore, it is to be understood that the inventions are not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.
