Q-VID: a revolutionary approach to vision-based measurements and control

Transcription

1 Q-VID: a revolutionary approach to vision-based measurements and control Ingrid Visentini 1, Riccardo M. G. Ferrari 1, Alessandro Ardesi 1 1 Danieli Automation SpA Via B. Stringher 4, 33042Buttrio (UD), Italy Phone: i.visentini@dca.it, r.ferrari@dca.it, a.ardesi@dca.it Keywords: computer vision for automation, vision-based instrumentation INTRODUCTION Vision-based applications are emerging as a new trend in the steelmaking industry, being exploited in a wide range of situations due to their advantageous tradeoff between deployment flexibility, cost and quality. Here we present a new instrumentation systems family, named Q-VID, capable of accurate measurements of hot and cold products by groundbreaking and robust computer vision techniques, specifically designed to fit accuracy and real-time constraints. We will describe its convenience with respect to more traditional, application-specific optical measuring approaches, and provide three real-world examples: an overview of a full system with tight deployment constraints, to show how the Q-VID modules can be deployed in out-ofreach plant infrastructure in a distributed fashion with no particular efforts; a case study for standard cameras chassis, to guarantee the components operability even in harsh conditions, such as high temperatures, dust and vibrations; a software framework, common to all the Q-VID products, which shapes several applications, and allows to meet the expected performance and reliability. VISION-BASED SYSTEMS: GUIDELINES Commodity vision systems are paving the road for new generations of instrumentation machines. Their low cost, easiness to maintain, and flexibility of application make them appealing for a wide range of applications, from simple area monitoring to more demanding tasks such as profile detection, surface inspection or robot guidance. Optic machines are already a well established technology to measure steelmaking products with timely and accurate responses; to give an example, the DANIELI HiProfile, which is a laser-based product, provides real-time non-contact profile shape inspection and dimension measurement with accuracy down to 0.015mm for a 1500mm of field of measure. However, albeit very accurate, they often come with a cost, which justifies their performances but discourage their usage in simpler and cheaper tasks. Moreover, their peculiar rigid geometry, which involves a fixed displacement of cameras and lasers, forces them to reside in ground plant areas and preclude them others. Even their intelligence is often represented by monolithic software with little possibility of expansion (let alone with custom components) and reuse. All these observations led to the creation of a family of products, named Q-VID, which is based on off-the-shelf high-resolution optical systems, protected with a custom-designed casing, suitable for contact-less on-line measurement and inspection of cold and hot products in the harsh environment of a steelmaking plant. It boasts modularity, reusability and versatility as main thread for assembling. Oppositely to stand-alone machines, with low opportunity of customization, we aimed at a flexible structure with room for expansion, scalability and interchangeability. Modularity has been seen as a key component and a primary goal [1]; thus the effort has been directed to create small and reusable components, both hardware and software, which can be combined to quickly generate a fullfledged application. Contrarily to prototyping, where performance is often sacrificed to the availability of a working system, here the basic modules are individual micro-systems that provide reliable and accurate workforce in a minimum time to market, in an attempt to build new products without being too disruptive with respect to existing hardware and software, but preserving the quality of the final result. The Q-VID lightweight hardware components are thought to be resistant, durable and self-containing elements. Each one of them is composed of easy-changeable parts, which ease the maintenance process and the repeatability. One of the most interesting aspects is their capability of being deployed in a variety of configurations, from ground level to roof mounting, in a distributed fashion or grouped together to achieve the maximum compactness. This feature opens a broader spectrum of possibilities, allowing the creation of a range of applications that work on inaccessible plant areas, i.e. for tracking the position of ladles, or semi-finished products. Their compact size eases the maintenance process, since the operators can easily access the individual components and replace them with minimum effort.

2 AN EXAMPLE OF SPACE-CONSTRAINED SYSTEM Thinking about a complete vision system, we often imagine a single machine that encompasses all the necessary instrumentation, and shields all the components in a single chassis. However, there may be the need for a distributed system, which involves components mounted in different areas, especially when the object to measure is too long or large, and cannot be framed with a single sensor; in this case, a distributed acquisition system may be more convenient, coupled with a separate central control system to process all the information in real-time. A practical example of the latter is the Q-VID ODL system, which measures the outer diameter (OD) and length (L) of seamless pipes after the piercing. The system is composed of four cameras, two dedicated to the OD and the remaining two to the L measurement. Figure 1: The OD+L system is composed of two main modules for measuring the Outer Diameter (left) and Length (right) of pipes; even they can be considered as separate systems, their outputs converge in a single processing unit. The outer diameter of a seamless pipe has to be measured at a nominal distance of 1.6 meters at the exit of a gage while the pipe is still being pierced. The two cameras are displacedd at roughly 90 degrees each other, as shown in (Figure 1), in a small space in between the piercer and a long array of lifting arms. These space constraints are precluding the possibility to install a single, voluminous gauge, but this represents the ideal situation for a distributed solution, flexible enough to allow a large tolerance in the positioning of the sensors. Figure 2: Positioning of the cameras for the pipe length measuring (shell measuring cameras), and the Outer Diameter (OD) cameras. For the length measurement, the maximum detectable pipe size is up to 10.5 meters at a distance of 4.5 meters in an environment that does not allow the installation of a bulky or heavy system (Figure 2). In fact, two high resolution cameras in sturdy, water cooled camera enclosure with fully integrated air purge, are placed on the top of a hood, and must offer minimum maintenance despite working in harsh environments. As we can see from Figure 3, the cameras can be placed above the lifting arms, far from each other;

3 even in this case, the configuration may slightly vary from design time, and the system must be adaptable to changes. For instance, in case of a longer pipe, the system can be extended with more cameras, which will not necessarily account the same spatial configuration of the previous ones. In both systems, the contact-free calibration step is adding another important and helpful featuree that ensures the operators will not have to retouch the enclosures position, allowing a fast and robust installation. Figure 3: Alternate view for the positioning of the cameras for the Length measuring (in orange, mounted on poles), and the Outer Diameter system cameras (in purple). A BASIC HARDWARE MODULE: CAMERA ENCLOSURE A trivial, but a fundamental issue to solve in designing an industrial camera system for steel-making plants, is to make the camera survive in the very harsh environment typically found in these premises. The main environmental aspects to consider are: High or low ambient temperatures (ranging from -20 to +60 C) High thermal radiation from hot steel products High level of vibrations High level of contaminants (e.g. dust) Apart from the obvious requirement of keeping the camera in a dust and water-tight enclosure (usually IP-67 or NEMA-4 rated), the issues of temperature, thermal radiation and vibrations must be especially dealt with. For this reason, the Q-VID family of cameras have been equipped with a specially designed, modular enclosure (see Figure 6and Figure 5), which are water cooled, in order to keep the camera at a constant suitable temperature by removing all the heat due to convection effect with the external air, and due to thermal radiation from hot products and entering the enclosure through the front optical window. This last aspect have been engineered with special care, as many applications do require a very close distance to hot steel products, such as the case of Figure 5 where a mounting distance of just 25 cm from a billet with temperatures in excess of 800 C is required. To correctly size the optical front window and the cooling requirement, an analytical computation of the radiated heat entering the enclosure has been carried on. First of all, it is useful to recall here the law governing the amount of heat radiated by a black or gray body (see ref. [2])

4 (1) where is the spectral radiant emittance of an ideal black body in an half sphere, per wavelength and per radiant surface area. The constants and and the formula were discovered by Planck, while is the wavelength and the temperature. A typical plot from eq. (1), for a temperature of 950 C, is drawn in Figure 4: spectral emittance of a black body radiator at 950 C. The peak is centered at about 2.5 microns.figure 4. Figure 4: spectral emittance of a black body radiator at 950 C. The peak is centered at about 2.5 microns. To get an idea of the huge amount of power involved, it suffices to consider that, by integrating (1) over all the wavelengths, it results that a body at a temperature of 950 C will emit about 104 KW per each square meter of its surface. It is of course clear that special care must be taken into shielding sensitive equipment such as cameras from this source of heat. Fortunately, a simple yet effective method of limiting the amount of heat entering the enclosure as thermal radiation is to limit the clear aperture through which the camera sees the hot steel product. For instance, in the speedmeter application shown here, thanks to the use of a sighting tube just a portion of about m 2 of hot product is directly in view of the camera. In order to compute the amount of heat, generated from such a small portion that enters through the Q-VID enclosure front window, we can use the following formula for computing the so-called view factor, which holds for a rectangular emitting and a rectangular receiving surface: & % % & % % (2) "#! "$ "# "$ & & (3) Eq. (2) defines the amount of heat emitted by one rectangle and which is absorbed by another rectangle, taking into account both the rectangles relative position and relative orientation, its terms being defined in the following drawing. 4 3 " 3 4 In our case, the first rectangle is the portion of hot steel product in sight, and the second rectangle is the optical window through which the camera, places inside the enclosure, can see the steel product. For the SM3200 case, it holds 7.025,10., so that the thermal power incident on the optical window reduces just to / 1 W which is a quantity several orders of magnitude smaller than the total power emitted by a typical cast product. This amount can be further reduced by using a heat absorbing glass for the front optical window, thus making the radiated heat entering the camera enclosure negligible, and leaving the camera own internal heat generation, and the heat exchange due to convection through the enclosure walls, as the main source of heat reaching the camera. Finally, suitable vibration dampening mounts have been installed, whose size and type has been designed by taking into account the camera system weight, and thus the resonance frequency of the system formed by the connection of the dampers and the enclosure, and the spectrum of the environmental vibrations at the chosen installation site.

5 Hot billet (between 800 and 1000 C) Measuring spot Q-VID camera under a protective cover Figure 5: example of a very demanding installation of a Q-VID camera (actually a SM3200 speedmeter), at about 25 cm from a billet at temperatures in excess of 800 C. Quick change front window Cooling water in/outlet Vibrationproof mounting a) b) Purge air inlet Laser pointer activation, status led and MIL electrical connector c) Figure 6: different views of existing industrial enclosures for the Q-VID family: a) a continuous caster SM3200 speedmeter, and b) and c) a camera module for Outer Diameter and Length measurements in pipe rolling mills.

6 ONE FRAMEWORK FITS ALL: THREE DIFFERENT APPLICATIONS WITH A LOT IN COMMON We already mentioned that the Q-VID VID family is aimed to cover a wide range of vision applications by the adaptation of basic components in flexible product development. The modularity and scalability of the system can be demonstrated by a concrete example of three main products,, which suit the most different requirements, and are composed of Q-VID VID basic blocks that allow cutting down their time-to-market without affecting the overall quality. Considering the following three vision applications, applications apparently they do not seem to have much in common: Q-VID CASTEYE:: positioned on the continuous casting line, close to the tundish, it monitors the meniscus area, computing the percentage of remaining powder in the crystallizer by image processing algorithms. Up to 10 independent cameras could be plugged to the controlling device and the system is required to work fluidly with hard real time constraints. Finally, the vision system may signal the overcoming of a tolerance threshold, to alert the operator, or to activate a robotic arm, arm to manage the powder [2]. The main effort in this case is directed to the handling of the sensors and the synchronous communication of results to the following operating devices. Figure 7: Q-VID VID CASTE CASTEYE working conditions (left) and input of the system (right). Q-VID WELD: its goal is to precisely control a welder machine, in order to align two sheets of metal prior to actual welding [4, 5].. The sheets are hold by two clamps, one static and the other moving in every direction. The vision system exploits two cameras and two light sources to acquire a set of images of two metal metal sheets and compute their corners displacement. displacement Figure 8:: Q-VID Q VID WELD design (left) and operating environment (right).

7 While computational time is not an issue, the accuracy of all corners positions has to be in the order of 0.25pixels, to guide the clamps for extremely precise welding. Q-VID ODL: as described in the previous sections, the system measures the outer diameter and length of seamless pipes while these are being processed by the piercer. The sensors exploit the IR emission of the material to acquire the presence of the pipe and must translate this information into a timely measure to the operator [6]. The acquisition frame rate is up to 250 fps, and the system must support the high throughput and perform the computation according to the real-time schedule; the required accuracy is 0.1% for the length and 5% for the diameter. The major effort in this case is represented by dealing with an intense acquisition pace, together with the need for a robust but lightweight measuring algorithm. Figure 9: Q-VID OD input images (left) and measured output visualization (right). Middleware DataBase HMI PLC L2 connection Operator Elaboration core (application-specific measuring system) L1 connection/ automation Ethercat Modbus Vision device Cameras 3D sensors Third party sensors Figure 10: Rough sketch of Q-VID logical architecture. Even if the purpose is different, they have a common architecture (Figure 10). In fact, all the Q-VID products rely on one or more cameras for acquiring images of a certain target and provide some measurements to operating workforce (human, robots or other automation parties). In particular, they share: I/O for communication with plant digital or analog devices: inputs/ results from/to PLCs or other automation devices. Grabbing from diverse optical devices: the acquisition module must support N independent sensors, concurrently or not. Publication of results on HMI or equivalent output channel: the measures must be processing, or to the HMI for visualization. sent to the automation for further Historical data storage: the output must be stored in a database, which may also be used for storing ancillary input data. The software structure is therefore organized in interchangeable modules to attain maximum system performance and development productivity. It is clear that the measuring principle varies for each application; this is the only software part that needs to be heavily revisited or completely changed for each product. The minimum requirements for all the software components are: reliability, realtime constraints satisfaction, configurability, simplicity, scalability.

8 CONCLUSIONS Born to aid the operators in common and repetitive monitoring tasks, vision systems are advancing as versatile, adaptable and independent applications dedicated to measuring, controlling, inspecting and guiding robotic arms. We here presented the Q-VID instrumentation systems family for accurate measurements of hot and cold products by computer vision techniques, designed to satisfy accuracy and real-time constraints according to specific application requirements. Its strengths are the capabilities to being built from minimal, independent, interchangeable modules; this fast build-up speeds up the product completion and minimizes the time to market. being deployed in a scattered, distributed manner in difficult plant areas, without affecting the overall performances. ACKNOWLEDGEMENTS The authors would like to thank their colleagues Marco Gobbesso and Mauro Linda for the fruitful support and discussions. REFERENCES 1. D. L. Parnas. On the criteria to be used in decomposing systems into modules. Commun. ACM, 15(12): , W. J Smith. Modern optical engineering. Tata McGraw-Hill Education, F. Della Vedova, Robotized system to manage the powders in a continuous casting plant for steel, US patent US , J. F. Lancaster, Metallurgy of Welding, 6th Edition. Elsevier, P. G. Davey, An Implementation of Model-Based Visual Feedback for Robot Arc Welding of Thin Sheet Steel. Int. J. of Robotics Research, Vol. 4, No 1, pages A. Horwitz, Ellipses of minimal area and of minimal eccentricity circumscribed about a convex quadrilateral. Australian Journal of Mathematical Analysis and Applications, Vol 7, No 1, Article 8, 2010.