Lance condition monitoring

Abstract

The present invention relates to a method (100) and a system (10) for monitoring the condition of an oxygen or instrument lance (5) in an oxygen blowing process for steel production. In this process, at least one image (7) of a lance (5) to be monitored is captured (S1) in an operating Linz-Donawitz converter, and a reference image (8) of the lance (5) to be monitored is loaded from a memory (3) (S2). The at least one captured image (7) is compared with the loaded reference image (8) (S3), and fault information is output to operating personnel based on the result of the comparison (S4).

Description

field of technology

[0001] The present invention relates to a method and a system for monitoring the condition of an oxygen or instrument lance in an oxygen blowing process for steel production. State of the art

[0002] Several processes are known for steel production. The so-called basic oxygen process (BOP), also known as the Linz-Donawitz process (LD process), is widely used. In this process, a converter, the so-called Linz-Donawitz converter (LD converter), is charged with liquid pig iron, a coolant, and a slag former, such as lime or dolomite. Pure oxygen is blown onto the feedstock through a usually extendable, water-cooled lance. Similarly, in argon oxygen decarburization (AOD), oxygen is blown into the molten steel from above in a corresponding converter using a lance.

[0003] Using various disposable measuring sensors, which are attached to a so-called sublance (instrument lance) via a contact system, process variables in the steel melt can be recorded, which are then used as the basis for controlling the converter process.

[0004] Due to the high temperatures of 1600°C prevailing in such converters, and the outgassing produced by the decarburization process and the combustion of some feedstocks, extremely harsh conditions exist within these converters. These inevitably lead to heavy stress and typically rapid aging of the lances used. Since some signs of aging and wear on the lances cannot be measured by sensors, manual inspection of the lances is usually necessary. Summary of the invention

[0005] Against this background, it is an object of the present invention to improve, and in particular to automate, the condition monitoring of oxygen or instrument lances in oxygen blowing processes for steel production in order to ensure plant safety and the productivity of the converter plant.

[0006] This problem is solved by a method and a system for monitoring the condition of an oxygen or instrument lance in an oxygen blowing process for steel production according to the independent claims.

[0007] Preferred embodiments are the subject of the dependent claims and the following description.

[0008] According to a first aspect of the invention, the method, particularly computer-implemented, for monitoring the condition of an oxygen or instrument lance in an oxygen blowing process for steel production comprises the following steps: i) sensorial acquisition of at least one image of a lance to be monitored in an operating converter for refining iron; ii) loading a reference image of the lance to be monitored from a memory and, in particular, automated comparison of the at least one acquired image with the loaded reference image; and iii) outputting fault information, preferably to operating personnel, based on a result of the comparison.

[0009] One aspect of the invention is based on the approach of comparing a current, preferably digital, image of a lance to be monitored in an LD converter with a, preferably digital, reference image of this lance. The reference image expediently shows the lance to be monitored in a normal state, i.e., an optimal state or a new state. The normal state thus preferably corresponds to a state in which the function of the lance is not impaired. The reference image may have been acquired at an earlier time and is expediently stored in memory.

[0010] The current image of the lance being monitored can be captured, for example, using a camera. The comparison of this captured image with the reference image is preferably automated. For example, the comparison can be performed by a suitably trained algorithm. This allows for reliable, precise, and—if new images of the lance are continuously captured—essentially continuous monitoring of the lance's condition. In particular, it is no longer necessary for the converter's operating personnel to devote their attention to monitoring the lance. Instead, the operating personnel can concentrate on other tasks, as error information can be generated based on the comparison of the captured image with the reference image. This error information can alert the operating personnel to the presence of a so-called...The error message may indicate an anomaly, for example, a deviation of the monitored lance's condition from its normal state. However, the error information may also characterize the anomaly detected through comparison.

[0011] For example, it is conceivable that error information is displayed when a deviation is detected between the captured image and the reference image and / or this deviation reaches or exceeds a predefined threshold. In this case, the operating personnel can be automatically alerted, for example, by a corresponding visual or audible alarm signal or by displaying a message on a graphical user interface of a control system for the converter. In particular, the error information can also be sent to operating personnel via a messaging service, such as SMS, email, MS Teams, WhatsApp, and / or similar methods.

[0012] A particular advantage of the proposed condition monitoring of the oxygen or instrument lance by comparing a captured image with a reference image is that geometric anomalies of the lance being monitored, such as plastic deformations or significant material loss, can also be detected. Such geometric anomalies cannot be detected, or can only be detected with considerable effort, in the usual automation data, i.e., sensor data such as temperature readings, concentration readings, pressure readings, and / or the like, which form the basis for process automation or control.

[0013] Preferred embodiments of the invention and their further developments are described below. These embodiments can be combined with each other and with the aspects of the invention described below, unless expressly excluded.

[0014] In a preferred embodiment, when comparing the at least one captured image with the reference image, it is checked whether a shape associated with the lance in the at least one captured image differs from a shape, particularly a geometric one, associated with the lance in the reference image. For example, it can be checked whether the contours of the lance in the captured image and the lance in the reference image are the same. Thus, the comparison allows, for example, checking whether these contours are congruent or can be aligned. More generally, the appearance of the lance in the at least one captured image and in the reference image can also be compared. Therefore, comparing the at least one captured image with the reference image offers a simple way to detect malfunctions or signs of aging and wear at an early stage.In particular, plastic deformations of the lance, such as bending, denting, material loss and / or material deposits on the lance, can be detected with particular reliability.

[0015] Error information can be displayed, in particular, as a deviation of the lance's shape in the captured image from the lance's shape in the reference image. Specifically, this deviation can be highlighted in one image, and this modified image can then be displayed as the error information.

[0016] In a further preferred embodiment, a measure of the deviation of the shape associated with the lance from the at least one acquired image from the shape associated with the lance from the reference image is determined, and the error information is output depending on the determined measure. The error information can be output, for example, when the determined measure reaches or exceeds a predetermined value. Determining such a measure of the shape deviation can allow for a more precise assessment of the existing anomaly, in particular its severity. The measure can thus indicate, for example, how far the detected anomaly has progressed. Optionally, such a measure can also be used as a basis for taking action, for example, if a sublance is to be bent back into its original shape by counter-movement or if an oxygen inflation lance is to be replaced with a new one.

[0017] In a further preferred embodiment, the deviation of the shape associated with the lance in the captured image from the shape associated with the lance in the reference image is highlighted in the captured image. Preferably, the resulting processed image is output, at least as part of the error information. In other words, post-processing of the captured image can be performed, providing the operator with visual feedback about the detected anomaly. This makes it easy for even less experienced operators to identify the problem or anomaly.

[0018] In a further preferred embodiment, the comparison determines whether and / or to what extent i) slag is adhering to the lance, ii) the lance's orientation has changed, iii) the lance is bent, iv) a lance head is worn, and / or coolant is leaking from a crack in the lance. The type of anomaly present can thus be determined by comparison. In other words, the detected anomaly can be categorized. In particular, the deviation of the recorded image from the reference image can be assigned to a specific error category. The aforementioned anomalies i) - v) correspond to the errors that typically or most frequently occur in a converter or oxygen blowing process. These typical anomalies can be reliably distinguished from one another based on the evaluation of the recorded image, particularly its geometric analysis.This distinction can then be usefully output, at least as part of the error information, so that the operating personnel can take targeted countermeasures. For example, the lance alignment can be corrected, the lance replaced or bent back, and / or a new lance head welded on.

[0019] In a further preferred embodiment, movement is detected in several captured images of the lance by comparison and used as the basis for the output of fault information. For example, coolant flowing along the lance can be detected, perhaps based on several successive comparisons of multiple captured images with the reference image and / or, optionally, also of the captured images with each other. A detected movement in the area of ​​the lance can therefore indicate cracks in the lance. Thus, a leak can be identified by motion detection. The specific search for movement in the images used for comparison allows for particularly reliable condition monitoring of the lance, since such movements very reliably indicate faults.

[0020] In a further preferred embodiment, a measure of the impairment of operational safety is determined based on the comparison and output, at least as part of the fault information. For example, the anomaly identified by the comparison can be assigned to the category "dangerous" or "non-dangerous," and this assignment can be output as part of the fault information. This significantly increases operational safety.

[0021] In a further preferred embodiment, the comparison is performed by a machine-learned model. A machine-learned model within the meaning of the invention is preferably a statistical model generated by one or more algorithms based on training data. The machine-learned model can therefore also be referred to as a trained model. Advantageously, the machine-learned model is the result of machine learning. Such a machine-learned model can also be commonly referred to as artificial intelligence, which has recognized patterns and regularities in the training data and can thus also assess or evaluate unknown data according to these patterns and regularities.

[0022] The machine-learned model can be trained using supervised learning. This is particularly advantageous if the images provided in a training dataset are essentially consistently annotated, i.e., the condition of the depicted lances is at least marked as "good" or "bad".

[0023] Alternatively, the model can be trained using semi-supervised learning. This is particularly advantageous if the provided images of a training dataset are only partially annotated and / or the dataset contains only a few images of lances in a "poor" condition. In this case, the model can be effectively retrained through continuous data collection using so-called pseudolabeling or entropy.

[0024] Finally, the model can also be trained using unsupervised learning. This is particularly advantageous when many images are available showing the lances in a "good" condition, meaning the training dataset contains no lance anomalies. For unsupervised learning, autoencoders with a convolutional neural network can be used, for example. Here, the input image is compared with a reconstruction (i.e., behind the decoder layer). If a large deviation is present, an anomaly can be assumed. Images with errors can thus be distinguished from normal images, and error categories (category 1, 2, ...) can be defined by subsequently clustering the features from the autoencoder's bottleneck layer (e.g., K-means, DBSCAN, hierarchical clustering, GMM (Gaussian Mixture Models, ...)).

[0025] According to a second aspect of the invention, the system for monitoring the condition of an oxygen or instrument lance in an oxygen blowing process for steel production, in particular for carrying out the process according to the first aspect of the invention, comprises: i) a sensor device for capturing at least one image of a lance to be monitored in an operating converter for refining iron; ii) a memory in which a reference image of the lance to be monitored is stored; iii) a comparison module which is configured to load the reference image from the memory and to compare at least one image captured by means of the sensor device with the loaded reference image, in particular automatically; and iv) an interface via which fault information based on a result of the comparison can be output, preferably to operating personnel.In particular, the comparison module can also be configured to generate the error information based on a result of the comparison and to initiate the output via the interface.

[0026] The system expediently includes a control device designed to execute the process steps of the procedures described herein.

[0027] Such a system allows for the reliable and essentially continuous collection of information on the condition of the monitored lance, without requiring specific attention from operating personnel. This enables earlier problem identification and thus the prevention of potentially catastrophic operational disruptions. In particular, lances in the harsh process environment within a converter can be monitored without exposing operating or maintenance personnel to the hazards posed by process conditions. This can also detect anomalies that cannot be identified by other sensors used to monitor converter process parameters.

[0028] A module according to the present invention can be configured as hardware and / or software. In particular, the module can comprise a processing unit, preferably connected to a storage and / or bus system via data or signals. For example, the module can comprise a microprocessor unit (CPU) or a module thereof and / or one or more programs or program modules. The module can be configured to execute instructions implemented as a program stored in a storage system, to acquire input signals from a data bus, and / or to output signals to a data bus. A storage system can comprise one or more, in particular different, storage media, especially optical, magnetic, solid-state, and / or other non-volatile media. The program can be configured such that it at least partially embodies the methods described herein.is capable of performing, so that the module can execute at least some of the steps of such procedures and thus, in particular, monitor an oxygen or instrument lance.

[0029] The sensor device is preferably designed as a camera. The camera can also be part of a conventional ITV (Industrial TV) system.

[0030] In a preferred embodiment, the sensor device is configured as a camera for detecting infrared radiation or as a laser scanner. By configuring the sensor device as an infrared camera, difficulties arising from the potentially poor lighting conditions in an LD converter can be circumvented. The same applies to the use of a laser scanner, which does not require active illumination of the lance to be detected.

[0031] In another preferred embodiment, the reference image is based on a previously captured image of the lance or on technical documentation. The reference image can, for example, be a CAD image or a similar design drawing of the lance, or one derived from it. This ensures that the reference image depicts an undisturbed state of the lance. Brief description of the drawings

[0032] The properties, features, and advantages of this invention described above, as well as the manner in which they are achieved, will become clearer and more readily understandable in connection with the following description of an exemplary embodiment, which is explained in more detail in conjunction with the drawings. These drawings show: FIG 1 shows an example of a system for monitoring the condition of an oxygen or instrument lance, and FIG 2 shows an example of condition abnormalities of oxygen or instrument lances.

[0033] Where appropriate, the same reference numerals are used in the figures for the same or corresponding elements of the invention. Description of the embodiments

[0034] FIG 1 Figure 1 shows an example of a system 10 for condition monitoring of an oxygen or instrument lance 5 in an oxygen blowing process for steel production. The oxygen or instrument lance 5 is arranged in a Linz-Donawitz converter (LD converter) or another converter for refining iron (not shown). The system 10 comprises a sensor device 1, a memory 3, a comparator module 2, and an interface 6. The system 10 is in FIG 1 This is purely an example of an automation system 4, for instance a programmable logic controller (PLC) for controlling the oxygen inflation process. System 10 can also be part of this automation system 4, if necessary.

[0035] The sensor device 1 is configured to capture at least one image 7 of the lance 5 to be monitored during operation of the LD converter. For this purpose, the sensor device 1, or at least a part of it, can be located inside the converter. Advantageously, the sensor device 1 comprises at least one optical sensor, for example, a camera or a laser scanner. Accordingly, the sensor device 1 can be configured to detect electromagnetic radiation with a wavelength in the visible and / or infrared range. Using appropriate optics, the sensor device 1 can thus capture the image 7 of the lance 5 to be monitored or generate it while scanning the lance 5. Optionally, the sensor device 1 can also provide a stream of multiple images 7, for example, to enable essentially continuous monitoring of the lance 5.

[0036] Interface 6 is used to output error information relating to lance 5 to operating personnel. Interface 6 can be implemented as either a hardware or software interface. For example, interface 6 could be a loudspeaker, a signal light, or a screen through which operating personnel can be informed of an anomaly. However, interface 6 can also be a software interface through which the error information can be output to other data processing devices, possibly for further processing. For example, such an interface 6 can be used to transmit the error information to the automation system 4, which can then take this information into account when controlling the converter process, i.e., the oxygen inflation process.Alternatively or additionally, the error information can also be displayed to the operating personnel on a graphical user interface of the automation system 4. It is therefore conceivable that the automation system 4 embodies or provides interface 6.

[0037] Memory 3 stores a reference image 8 of the lance 5 to be monitored. Reference image 8 conveniently shows the lance 5 in a normal state, i.e., a state in which there is no fault and the function of the lance 5 is not impaired.

[0038] The comparison module 2 is configured to load the reference image 8 of the lance 5 to be monitored from memory 3 and to compare the at least one image 7 recorded by the sensor device 1 with the loaded reference image 8. Based on this comparison, the comparison module 2 can then initiate the output of the error information via interface 6, for example, if the appearance of the lance 5 in the recorded image 7 differs from the appearance of the lance 5 in the reference image 8.

[0039] The comparison module 2 preferably comprises a machine-learned model 9. The comparison module 2 thus preferably comprises an algorithm that detects an anomaly of the lance 5 in the at least one recorded image 7 by detecting unusual patterns or outliers compared to standard data, as embodied by the reference image 8. This machine-learned model 9 is preferably designed as a convolutional neural network.

[0040] Comparison module 2, or at least its algorithm, in particular the machine-learned model 9, is expediently trained to identify these unusual patterns or outliers in the at least one captured image 7. This training is expediently carried out on a training dataset of standard images or videos. The identification of features (so-called features such as edges, corners, textures, and colors) that embody the patterns or outliers can be manually extracted from these standard images or videos and used as part of the training dataset for the learning process, i.e., the training of model 9. However, it is also conceivable that model 9 learns such features or the underlying hierarchy automatically from the raw data, for example, from the pixels of a large number of images.

[0041] Therefore, training Model 9 can involve unsupervised learning, especially when data with manually extracted features associated with anomalies are unavailable or only available to a limited extent. In this case, Model 9 can recognize a distribution of the data and mark deviations or outliers from this distribution as anomalies. If, on the other hand, a sufficient amount of data is available in which the features or their associated anomalies have been manually extracted or marked, the model can explicitly distinguish between normal and abnormal patterns; this is referred to as supervised learning. A combination of these two learning methods can also be used, for example, when a small amount of data with marked anomalies and a large amount of unmarked data are available.

[0042] In principle, the comparison module can employ two different techniques for anomaly detection. One possibility is reconstruction-based techniques, where the corresponding machine-learned model 9 is trained to reconstruct the reference image 8, i.e., an image in which the monitored lance 5 is in a normal state. The reconstruction errors then indicate the anomalies. Such reconstruction-based techniques are used, for example, by so-called autoencoders or generating adversarial networks. Another approach is to train the model 9 to draw a boundary for deviations from norm data, as characterized by the reference image 8. This allows the model 9 to classify everything outside this boundary as an anomaly. Such techniques are used, for example, by vector machines. Finally, statistical and distance-based techniques can also be employed.Model 9 can therefore also be trained to measure deviations of new data points from the recorded images 7 from a norm distribution learned during training. Large deviations can then be flagged as anomalies.

[0043] The comparison module 2 can also be configured to employ post-processing techniques. Techniques such as localization using saliency maps or other attention mechanisms are expediently used. This allows the comparison module 2 to mark or highlight those areas in the at least one captured image 7 where the anomaly was detected.

[0044] Using system 10, a method 100 for monitoring the condition of the oxygen or instrument lance 5 in the oxygen blowing process for steel production can be implemented. In a process step S1, the sensor device 1 records at least one image 7 of the lance 5 to be monitored. This image is expediently provided to the comparison module 2. In a further process step S2, the reference image 8 is loaded from memory 3 and is also expediently provided to the comparison module 2. In a further process step S3, the at least one recorded image 7 is compared with the reference image 8. Based on the result of this comparison, the fault information can then be output to the operating personnel via interface 6 in a further process step S4. In particular, the comparison result itself can also be output as fault information.

[0045] FIG 2 Figure 5 shows an example of various anomalies of an oxygen or instrument lance 5 that can occur during operation in a Linz-Donawitz converter. These anomalies can, for example, result from the harsh environmental conditions inside this LD converter.

[0046] FIG 2A Figure 5 shows slag 11 baked onto the lance 5 in the area of ​​a lance head 5a. Such baking of slag 11 can lead to the lance 5 no longer being able to be extended out of the LD converter through its corresponding access opening, or at least not without causing damage.

[0047] FIG 2B Figure 5 shows a misalignment of lance 5 by an angle α, caused, for example, by the influence of high temperatures. Such a misalignment also prevents lance 5 from being easily inserted through an access opening 12 of the LD converter.

[0048] The same problem arises for the in FIG 2C depicted, bent lance 5.

[0049] FIG 2D shows a lance 5 whose lance head 5a is worn, for example corroded, which may impair the functionality of the lance 5.

[0050] FIG 2E Figure 5 shows a lance 5 in which several cracks 13 have formed. During operation of this lance 5, coolant can escape from the cracks 13, which means that sufficient cooling of the lance 5 can no longer be guaranteed.

[0051] A comparison module of a system for condition monitoring of an oxygen or instrument lance 5, as is used, for example, in FIG 1 The system shown can be configured to determine, by comparing a captured image of the monitored lance 5 with a reference image of lance 5, whether one of the anomalies shown here is present. The anomalies from FIG 2A, 2B, 2C und 2D For example, they can be detected by comparing the appearance of lance 5 from the at least one recorded image with the appearance of this lance 5 in the reference image. In particular, the shape associated with the depicted lance 5 can be compared with the shape of lance 5 from the reference image. This is because the anomalies from FIG 2A, 2B, 2C und 2D They are essentially geometric in nature. A suitably configured comparison module can therefore easily distinguish the various anomalies from one another and categorize them accordingly.

[0052] To also the in FIG 2ETo detect the crack formation shown in lance 5, the comparison module is expediently configured to detect movement in the area of ​​the monitored lance 5 based on multiple comparisons of several recorded images with the reference image, or, if necessary, also of the several recorded images with each other. Such movement is to be expected in the case of crack formation caused by the leakage of cooling water.

[0053] Although the invention has been further illustrated and described in detail by the preferred embodiments, the invention is not limited by the disclosed examples and other variations can be derived from them by the person skilled in the art without leaving the scope of protection of the invention. Reference symbol list

[0054] 1 Sensor device 2 Comparison module 3 Memory 4 Automation system 5 Oxygen or instrument lance to be monitored 5a Lance head 6 Interface 7 Captured image 8 Reference image 9 Machine-learned model 10 System 11 Slag 12 Access opening 13 Crack 100 Procedures S1 Capture an image S2 Load a reference image S3 Compare S4 Output α angle

Claims

1. Method (100) for condition monitoring of an oxygen or instrument lance (5) in an oxygen blowing process for steel production, wherein - at least one image (7) of a lance (5) to be monitored is sensorially recorded in an operating converter for refining iron (S1), - a reference image (8) of the lance (5) to be monitored is loaded from a memory (3) (S2) and the at least one recorded image (7) is compared with the loaded reference image (8) (S3), - based on a result of the comparison, fault information is output (S4).

2. Method (100) according to claim 1, wherein when comparing the at least one recorded image (7) with the reference image (8) it is checked whether a shape associated with the lance (5) from the at least one recorded image (7) differs from a shape associated with the lance (5) from the reference image (8).

3. Method (100) according to claim 2, wherein a measure for the deviation of the shape associated with the lance (5) from the at least one recorded image (7) from the shape associated with the lance (5) from the reference image (8) is determined and the error information is output depending on the determined measure.

4. Method (100) according to one of claims 2 or 3, wherein the deviation of the shape associated with the lance (5) from the recorded image (7) from the shape associated with the lance (5) from the reference image (8) is marked in the recorded image (7) and the processed image thus created is output at least as part of the error information.

5. Method (100) according to one of the preceding claims, wherein, on the basis of comparison, it is determined whether and / or to what extent - slag (11) is baked onto the lance (5), - the orientation of the lance (5) has changed, - the lance (5) is bent, - a lance head (5a) is worn, and / or - a coolant is leaking from a crack (13) in the lance (5).

6. Method (100) according to one of the preceding claims, wherein a movement in several recorded images (7) of the lance (5) is determined by comparison and the output of the error information is based on this.

7. Method (100) according to one of the preceding claims, wherein, on the basis of the comparison, a measure of the impairment of operational safety is determined and output at least as part of the fault information.

8. Method (100) according to any of the preceding claims, wherein the comparison is performed by a machine-learned model (9) based on a convolutional neural network.

9. System (10) for monitoring the condition of an oxygen or instrument lance (5) in an oxygen blowing process for steel production, comprising: - a sensor device (1) for recording at least one image (7) of a lance (5) to be monitored in an operating converter for refining iron, - a memory (3) in which a reference image (8) of the lance (5) to be monitored is stored, - a comparison module (2) which is configured to load the reference image (8) from the memory (3) and to compare at least one image (7) recorded by means of the sensor device (1) with the loaded reference image (8), - an interface (6) via which fault information can be output based on a result of the comparison.

10. System (10) according to claim 9, wherein the sensor device (1) is configured as a camera for detecting infrared radiation or as a laser scanner.

11. System (10) according to claim 9 or 10, wherein the reference image (8) is based on a previously recorded image or technical documentation of the lance (5) using a camera.

Images