DE4040222A1 - Path control for object handling robots - is carried out by flexible program based on analysis of optical data - Google Patents

Abstract

The control provides decision criteria for sequential working system, using a system of electronic components. It generates a working process starting targetted actions according to optically recognised markings, providing electroptical signals for computerised control programs from object to object. Identified objects are tested and optical points of gravity detected for contour scanning. The obtained signals are computer processed for final checking, and the robot targetted actions initiated and controlled. USE/ADVANTAGE - For industrial processes, e.g. car body welding, with object identity determination and checking.

Description

Path controls of robots are general known, usually the path control in fixed predetermined paths are carried out, e.g. B. when leaving Spot welding sequences on bodies corresponding to one predetermined program flow. Such a program only allows exact repetitions of the given Path sequences, with corrections, for example to Compensate for a linear displacement between objects and robots plus / minus to the right or left to make a difference in position between Compensate robot and body to be machined.

Such known controls of robots are not able to process thought processes or differentiation or To carry out selection processes, for example Differentiate objects according to object classes. As a result, human workers can get on Recognition of selection criteria based work such as Sorting and / or correcting the position of objects use of robots with the known Path controls cannot be replaced.  

In many cases, however, it is desirable with regard to a humanization of corresponding jobs as well with regard to the absolute required Compliance with an error-free function, e.g. B. at the Sorting and / or handling objects corresponding workstations very much simple, but also dull handling relieve. Such manipulations also have the Disadvantage that they are the people involved make indifferent and thus mistakes challenge.

The invention is based on the object Process and a system for path control of To specify robots when handling objects with who succeed in finding individual objects according to their Belonging to a certain object class identify and for example not identifiable objects from the rest Eliminate proceedings, the situation of a Object in a scene to be analyzed to be determined and corrected on a case by case basis, and finally one of the standard non-uniformities deviating object, e.g. B. with sprues, strong Burrs, breakouts and other defects recognize and discard or align and to this purpose in accordance with such information initiate targeted actions by the robot.

The problem is solved with a method of type mentioned with the invention in that the path control by means of a  optically recognizable features of the objects and / or an optical detection of their position in one analyzing scene from object to object flexible calculable or changeable control program is made.

The advantage of the method is that individual objects identified by their class and unidentifiable from the rest Processing can be excluded that the position of an object in the scene to be analyzed for the purpose of correction by the robot at its Path control is taken into account and that objects with discernible irregularities or Flaws or with incorrect training by corresponding instruction to the robot Control program also from further Processing step excluded and for this purpose the robot or robots by a flexible control Providing one from the information from object to Object calculable control program in different and individually flexibly variable Movements are controllable.

An embodiment of the method according to the invention provides a sequence of work steps as follows:

• a) Feed individual objects from a supply a first detection station;
• b) Generation of several light sections in sequence for Extraction of features of individual objects and Extraction of their spatial as well  contrast-independent visual representations for Identification;
• c) Not discard and / or return identifiable objects from the rest Procedure;
• d) transfer identified objects to a second Detection or control station to determine optical focal points, and based on that, scanning relevant contours as well as for comparison with an or multiple threshold (s);
• e) processing the information obtained by a Arithmetic operation using a neural network to decide on the affiliation of an object to a known object class, as well as determining the Position of the object within the to be analyzed Scene;
• f) initiation of targeted actions for path control of the Robots by controlling them according to one the information b) to d) Control program.

Advantageous further developments of the invention Procedures are related to the features of the subclaims remove.

A system for path control of robots at Handling objects, especially for execution of the process, is characterized by an Functional unit formed by the following elements:

• - A device with means for fully automatic Feeding individual objects to a first one  Detection station;
• - A first detection station with means for contrast-free 3-D imaging of individual objects as well for electronic image evaluation for their Identifiability according to comparable, predefined pattern;
• - a facility with means for return or Discarding unidentifiable objects;
• - a first robot to grasp the recognized ones Objects and to drop onto a second detection or control station;
• - A second detection or control station with Means for checking the contour of the objects (e.g. Diameter, hexagon, position of a sprue, Flaws, etc);
• - a second robot for transporting and Align the identified and error-free Objects to a transfer station;
• - a handling system for inserting the objects in a Processing machine.

Further advantageous refinements of the system for Path control are in accordance with the subclaims intended.

The invention is shown in a schematic drawing in a preferred embodiment shown, from the drawing further advantageous details of the Invention are removable.

The system essentially has a first robot ( 1 ), a second robot ( 2 ) and the detection station ( 3 ) and the control station ( 4 ). All stations and an electrical cabinet ( 5 ) are accessible from the operating side ( 6 ), in particular for checking and retrofitting.

The objects or workpieces ( 10 ) are fed with a fully automatic feeder, consisting of an elevated container ( 11 ) with an outlet chute ( 12 ) and a linear conveyor rail ( 13 ), preferably a vibrating rail with coarse sorting, driven by a vibration pot ( 14 ) onto a small conveyor belt ( 15 ) and thus arrive at the first detection station ( 3 ).

The objects stored in the area of the first detection station ( 3 ) are checked there by means of optoelectronic part detection and testing according to belonging to an intended object class by comparison with predetermined comparison patterns according to certain procedural sequences (pattern). Objects that cannot be identified are pushed by the computer onto the return conveyor belt ( 16 ) by means of a corresponding path control - in some cases also by a separate sliding mechanism - and are preferably transported back into the elevated container ( 11 ) by the latter. Occasionally, the reject container ( 17 ) could also be separated. In this case, the return conveyor belt ( 16 ) would be equipped with a switch to separate out the objects.

The identified objects are taken by the robot ( 1 ) for contrast-invariant 3-D imaging and their evaluation and transported to the second recognition or control station ( 4 ) and stored there. This station is also a detection system with a camera and light source. Here, a contour check is carried out, for example, with regard to the spatial shape of the object, such as, for. B. hexagon or presence and location, for example a sprue or the presence of other faults such as burrs etc. Detected as faultless parts are picked up again by the robot ( 2 ) and transported to the transfer station ( 18 ). There they are stored at three predefined transfer points. Objects identified as defective are pushed to the reject container ( 17 ) by separate sliders.

The elevated tank ( 11 ) with z. B. 600 liter bunker volume is preferably lined with plastic and equipped with a bridge breaker. The outlet chute ( 12 ) is designed as a vibration channel and conveys the workpieces or objects ( 10 ) to the linear conveyor ( 13 ). The infinitely adjustable linear conveyor with rail and jam limit switch for controlling the discharge chute ( 12 ) is used for stock formation and alignment of the parts up to the detection station ( 3 ) designed as an image processing station. This consists of a controlled small conveyor belt ( 15 ) or a tiltable plate, equipped with a limit switch for switching off the linear conveyor ( 13 ).

In the second detection or control station ( 4 ), the projection of the object outline is read into the computer by means of a second CCD camera, as a result of which, as explained above, an assessment is made, for example, of the hexagon position, a sprue or an inner diameter or other features. Fault-free parts are then controlled and taken over or aligned by the second robot ( 2 ) and placed in a transfer station ( 18 ). A handling system ( 19 ) with a triple gripper feeds three parts depending on the cycle from the transfer station ( 18 ) to the further processing machine ( 20 ) or its clamping unit ( 21 ). The processing machine ( 20 ) is, for example, a thread cutting machine, the finished parts being conveyed into a finished product container ( 22 ).

The image processing system of the recognition station ( 3 ) consists of an imaging part and an image processing computer. A light beam is generated by the fringe projector ( 30 ), with the aid of which defined strips of light are cast onto an object or workpiece. These stripes are shifted in a defined manner by phase shift on the objects. Three or four of these phase-shifted images are read into a computer using the CCD camera. This creates a 3-D contour that is independent of the object. Any changes in contrast caused by surface impairments such as moisture, galvanizing, dust etc. play no role here. On the basis of the result of the first detection station ( 3 ), identifiable and correct parts in the tapping field ( 31 ) are taken over by the first robot ( 1 ) and transported to the second detection or control station ( 4 ). Unidentifiable parts are discarded as rejects ( 17 ) or conveyed back into the container ( 11 ).

The operation (display, input and output) takes place via an AT-compatible computer with screen, keyboard and mass storage (floppy disk + hard disk), which to the Image processing computer is connected. In order to can also be taught different parts that Save program information and save it later Get changeovers quickly. For this purpose only the pliers of the grippers can be replaced.

The image processing computer (not shown) consists of a transputer network with:

• - A master computer for image digitization;
• - three identical CPU cards for image analysis;
• - a control card for the Fringe projector and
• - two control units for the two robots.

The individual modules are designed so that all arithmetic operations including the additional control task performed in real time can be.

The classification is in both Image processing stations achieved in that the current input scene with taught reference scenes is compared. However, since fewer reference scenes than Actually occurring scenes exist neural network for generalizing the a priori  Information used.

The threshold value operation for 3-D information carried out in one of the detection stations ( 3 ) or ( 4 ) with the formation of related components in order to determine the optical focal points of the objects is based on the following findings and measures:
The representation obtained from the light section method is comparable to a map on which the respective height can be read at any point by means of a corresponding color. If you are only interested in areas that are higher than a certain height (threshold value), then you assign all points that are smaller with 0 (zero), all others with 1 (one). The areas ("CLUSTER") which are filled with 1 (one) are then the searched areas, in general the objects.

The algorithm below has the following Tasks:

• - Summarize all related "1-Pixels" Clusters;
• - filtering out "noise";
• - Calculation of noise invariant information about the location of the cluster (optical focus) in the scene to be analyzed.

As a result, the algorithm returns a lot of Points (pixels) at which an object (CLUSTER) is recognizable.  

When using neural networks for Object classification is carried out as follows: Since neural networks are interconnected, interacting "trainable" Are functional units, they are able Feature vectors from the feature space to taught Map classes. The programming takes place by "training", the implementation in one Computer program massive parallel programming allows. For this, the feature vectors are "Scanning" the surroundings of a "suspected object" won, which is an element of the object set acts as a result of the threshold operation emerges.

The neural network that previously had similar patterns "trained" is now able to decide which object class is the special one Object.

Analyzing object contours can be very beneficial for example using a hexagon determination are explained. The goals are:

• a) Determination of the position of the cast hexagon (Angle of rotation);
• b) detection of the position of a sprue;
• c) Classification of dimensional accuracy.

In all three cases, the work to be done is pure Classification, therefore, for this purpose also use neural networks.  

For the extraction of features, the procedure is such that the objects to be checked preferably on a light table illuminated from below and the silhouette with a Camera is recorded. The resulting Binary image serves as the following procedure Basis.

From the center of the test object, similar to the RADAR, the information is recorded along a beam, e.g. B. according to FIG. 2nd

If a sufficient number of such beams are plotted and the information is recorded in a linearized manner, the following image is obtained in accordance with FIG. 3.

In this way of representation, the patterns deliver different angle of rotation 0 to 360 degrees same pattern picture, but by a certain amount shifted to the right or to the left. The conclusion from this is that for the classification of an angle of rotation only the detection of the displacement vector a pattern is required, from which the Angle can be calculated.

Errors and irregularities in the object are like recognized as follows: A sprue, for example, manifests itself in Form of a pattern disturbance. The neural network is now trained so that it can determine in which sector the gate is located. From this information can in the control program for the robot intended rotation to the intended corrective alignment of the object is calculated and  are de facto derived.

The classification of the dimensional accuracy of an object can also be connected to a neural network with the above Patterns are performed using only two classes "good" / "bad" exist.

Claims (10)

1. A method for controlling the path of robots when handling objects, characterized in that the path control can be flexibly calculated or can be calculated from object to object by means of an optically recognizable features of the objects and / or an optical detection of their position in a scene to be analyzed changeable control program is made. 2. The method according to claim 1, characterized by the sequence of work steps:

• a) feeding individual objects from a supply to a first detection station;
• b) generating a plurality of light sections in sequence for extracting features of individual objects and obtaining their spatial and contrast-independent visual representations for identification;
• c) separating and / or conveying back unidentifiable objects from the further procedure;
• d) transfer of the identified objects to a second detection or control station for determining optical focal points and, starting from this, scanning relevant contours and for comparison with one or more threshold value (s);
• e) processing of the information obtained by a computing operation using a neural network to decide whether an object belongs to a known object class and determine the position of the object within the scene to be analyzed;
• f) initiation of targeted actions for path control of the robots by controlling them in accordance with a control program calculated from the information.

3. The method according to claim 1 or 2, characterized characterized in that for generating the light sections optical means such as interferometers, piezo-controlled Mirror or moving slides used and by these Streaks of light cast on the scene to be analyzed be, with at least three, preferably four or multiple pictures by the light strips each relative by a predetermined amount shifted towards each other, an information value about the Z coordinate (height) of each one Pixel is calculated. 4. The method according to one or more of claims 1 to 3, characterized in that the calculation of the Z coordinate according to the formula is carried out with
I / 1 = intensity at point (x, y) in Figure 1
I / 2 = intensity at point (x, y) in Figure 2
I / 3 = intensity at point (x, y) in Figure 3
I / 4 = intensity at point (x, y) in Figure 4
means. 5. The method according to one or more of claims 1 to 4, characterized in that when performing the threshold value operation image pixels of all object parts above a predetermined height (threshold value) as 1 (one) and all object parts below the threshold value as 0 (zero) and that an algorithm performs arithmetic operations with the following arithmetic steps:

• - Combining all related "1-pixels" into clusters;
• - filtering out "noise",
• - Calculation of noise-invariant information about the location of the cluster (optical focus).

6. The method according to one or more of claims 1 to 5, characterized in that the environment of a "suspected" object, i.e. H. of an object one Object quantity optically scanned and in the process Feature vectors obtained and this by using neural network trained through a similar pattern comparative mapping to taught classes are classified (step d) of claim 2). 7. System for path control of robots when handling objects, in particular for carrying out the method according to one or more of claims 1 to 6, characterized by a functional unit formed from the following elements:

• - A device with means ( 11 , 13 , 15 ) for the fully automatic feeding of individual objects ( 10 ) to a detection station ( 3 );
• - A first detection station ( 3 ) with means ( 30 ) for contrast-free 3-D imaging of individual objects and for computational electronic image evaluation for their identifiability in accordance with comparable, predetermined patterns;
• - A device with means ( 16 ) for returning or discarding unidentifiable objects;
• - a first robot ( 1 ) for gripping the recognized objects and for placing them on a second detection or control station ( 4 );
• - a second detection or control station ( 4 ) with means for checking the contour of the objects (e.g. diameter, hexagon, position of a sprue, defects, etc.);
• - a second robot ( 2 ) for transporting and aligning the identified and error-free objects onto a transfer station ( 18 );
• - A handling system ( 19 ) for inserting the objects in a processing machine ( 20 ).

8. Plant according to claim 7, characterized in that the device for feeding the objects ( 10 ) comprises an elevated container ( 11 ) with discharge chute ( 12 ) and a linear conveyor rail ( 13 ) and a small conveyor belt ( 15 ). 9. Plant according to claim 7, characterized in that the first detection station ( 3 ) comprises an image recognition system with a camera and a light source. 10. System according to one or more of claims 7 to 9, characterized in that the image recognition system comprises a fringe projector ( 30 ) as a light source, for optically recording an object, a CCD camera and for evaluating an image processing computer that the second recognition or control station ( 4 ) has a second CCD camera with light source and an image processing computer and a second robot ( 2 ), and that the handling system ( 19 ) has a triple gripper.