Method and non-contact distance sensor for securing a machine

Abstract

Method for securing a machine (10), in particular a robot arm, in which objects (18) in a working area (20, 22) of the machine (10) are detected by at least one non-contact distance sensor (12) moving with the machine (10) with at least one distance measuring beam (14) measures an actual measured value for a distance to the working area (20, 22) or an object (18) located there, a first virtual distance measurement beam (28) is formed from a pose of the distance sensor (12), which simulates the distance measurement beam (14), the point of impact of the first virtual distance measurement beam (28) on a surface model of the working area (20, 22) is determined, and a first virtual measurement value for the distance is calculated. and the actual measurement is compared with the first virtual measurement to decide whether an object (18) is located in the distance measuring beam (14), characterized by that at least one further virtual distance measurement beam (30) is formed with an offset (32) to the first virtual distance measurement beam (28), the point of impact of the at least one further virtual distance measurement beam (30) on the surface model is determined and at least one further virtual measurement value for the distance is calculated, that it is compared whether the actual measured value is compatible with the virtual measured values, in particular whether it is equal to at least one of the virtual measured values ​​within a tolerance (38), and that an object (18) is detected in the distance measuring beam (14) if the actual measured value is not compatible with the virtual measured values.

Description

[0001] The invention relates to a method and a non-contact distance sensor for securing a machine according to the preamble of claim 1 and 15 respectively.

[0002] The primary goal of safety engineering is to protect people from hazards, such as those posed by machinery in industrial environments. The machine is monitored using sensors, and if a situation arises in which a person is at risk of getting dangerously close to the machine, an appropriate safety measure is implemented. Sensors used in safety engineering must operate with exceptional reliability and therefore meet stringent safety requirements, such as the EN 13849 standard for machine safety and the IEC 61496 or EN 61496 standard for non-contact protective devices. To meet these safety standards, a number of measures must be taken, such as reliable electronic evaluation through redundant, diverse electronics, functional monitoring, and monitoring for contamination of optical components.

[0003] In the safety monitoring of robots, especially lightweight robots, there is a growing desire for closer collaboration with humans (HRC, human-robot collaboration), even in complex environments. Relevant standards in this context include ISO 10218 for industrial robots and ISO 15066 for collaborative robots. Similar challenges arise for machines other than robots, such as AGVs / AGCs (Automated Guided Vehicles / Containers) or drones.

[0004] With robots, a conventional approach limits the force to such an extent that even a collision cannot lead to serious injuries. Furthermore, the design of the robot's mechanical components deliberately focuses on smooth, rounded, or soft outer contours. However, this leaves potential hazards in the tool itself, which may have sharp, rotating, or hot parts. Moreover, these mechanical measures and force limitations do not prevent, for example, hand crushing injuries between the tool and a holder or similar object on a work surface.

[0005] It is now possible, in principle, to move a small, lightweight distance sensor, or multiple such sensors, along with the tool on the robot arm. DE 10 2015 112 656 A1 addresses such a distance sensor. This creates a kind of virtual protective bell or shield around the tool, and the robot is slowed or stopped if a safety-relevant object penetrates it. The distance sensor, in conjunction with the higher-level control system, is designed to be safe, and tests in the form of a reference run with known reference targets are possible to achieve a higher level of safety. Approaching surfaces must be individually programmed for each position to prevent the protective bell from being triggered by the surface.

[0006] Near-surface motion paths, such as those used to create an adhesive bead or a weld bead, cannot be implemented in this way. The conventional approach is also unsuitable for dynamic motion behavior, such as the unpredictable motion path during an evasive maneuver or a motion path in response to pose detection of a workpiece being gripped.

[0007] The EP 3 988 256 A1 provides a more flexible protective dome by adjusting distance thresholds, which effectively define the length of the measuring beams. This adjustment incorporates the maximum available distance to the nearest background object. The maximum distance is calculated from the expected path of the monitoring sensor's line of sight and the surrounding topography. However, problems arise when the assumed line of sight deviates from the actual line of sight, or when the assumed topography deviates from the actual surroundings. This is particularly critical at edge detection points, where even slight deviations can lead to significantly different distances.

[0008] DE 10 2018 118 265 A1 discloses a method and a monitoring system for safeguarding a machine, in which, during a learning phase, a distance-measuring sensor detects the environment of a robot and generates a reference map from this data. During an operational phase, the environment detected by the sensor is compared with the reference map, and a safety assessment of deviations determines whether a safety-related response is required. The deviation assessment is based on a correlation measure between the target view and the detected view, and in this general form, it is not suitable for safeguarding of the type described above, which is based on specific distance-measuring beams.

[0009] EP 3 112 900 A1 discloses a system for securing a machine with a distance sensor mounted on the machine. The measurement data is compared with a model of the expected background, which is formed, for example, by walls, excavated earth, and similar expected environmental objects. If the volume of a deviation would leave enough space for a person, the machine is secured. In some embodiments, several distance sensors are provided, for example, two distance sensors for redundant measurement and evaluation.

[0010] EP 4 385 676 A1 discloses a method for validating the pose of a robot and / or sensor data from a sensor moving with the robot. The robot's movements are compared with those of a robot model, and the sensor measurements are compared with those of a sensor model interacting with an environment model. As already discussed in relation to EP 3 988 256 A1, even slight deviations between the model and reality, especially at edges, can lead to large discrepancies between predicted and actual distances. This results in false-positive object detections, which significantly impair availability. In other words, productivity is lost due to safety measures taken for a safety-relevant object that was incorrectly identified as such based on inaccurate model predictions.

[0011] The purpose of the invention is therefore to enable more accurate object recognition.

[0012] This problem is solved by a method and a non-contact distance sensor for safeguarding a machine according to claim 1 and 15, respectively. The machine is, in particular, a robot or robot arm with a tool that is primarily to be safeguarded (EOAS, End-of-Arm Safeguarding). This application example of a robot is used frequently below as a representative example; the explanations can be applied analogously to other moving machines, such as a vehicle or AGV (Automated Guided Vehicle). Safeguarding is based on object detection within a workspace of the machine. The workspace itself is known in the form of an environment model or surface model, which can also be referred to as the topography or 3D contour of the workspace.Object recognition reacts to those objects that are unexpected, therefore do not belong to the workspace and are not represented in the surface model, and it primarily targets people or body parts to prevent accidents with injuries.

[0013] At least one non-contact distance sensor is moved along with the machine, preferably attached to it, and uses at least one distance measuring beam to determine an actual measured distance. Depending on whether a detectable object is located within the distance measuring beam, this measurement is either the distance to the work area or to the object itself.

[0014] From the position of the distance sensor, a first virtual distance measurement beam is generated, which corresponds to, models, or simulates the actual distance measurement beam. The term "pose" refers to the position and / or orientation in up to six degrees of freedom. The position of the distance sensor results from the position of the machine and any potential, predeterminable, and therefore correctable offset of the distance sensor relative to the machine. The machine's position can be determined, in particular, from the machine's forward kinematics. Forward kinematics is a term from robotics that can be applied to other machines. It refers to the determination of the position of an end effector, and possibly other kinematic parameters such as velocity and acceleration, from information about the moving elements—in the case of a robot arm, the joints, for example, from rotary encoders mounted there.In a robot, the forward kinematics are generally determined for its workflow anyway. This information can be used to derive the pose of the distance sensor. Within the accuracy limits of the pose determination, the position of the distance sensor clearly indicates the path of its distance measuring beam, allowing this to be simulated by the first virtual distance measuring beam.

[0015] From the point of impact of the first virtual distance measuring beam on the surface model of the work area, a first virtual measurement value for the distance is calculated. This is a prediction of the actual measurement value based on the first virtual distance measuring beam and the surface model, which together simulate the measurement behavior of the distance sensor in the current situation, provided no object is located in the distance measuring beam. For object detection, the actual measurement value is compared with the first virtual measurement value. Conventionally, a deviation beyond a certain tolerance would already indicate that an object has been detected. The invention employs an extended approach, explained below, in which the aforementioned comparison is not the sole determining factor.

[0016] The invention is based on the fundamental idea of ​​creating at least one additional virtual distance measuring beam and including it in the comparison. This additional virtual distance measuring beam is offset relative to the first virtual distance measuring beam. Analogously, the point of impact on the surface model is determined, and a further virtual distance measurement is calculated from this. The offset corresponds intuitively to a possible deviation in the predicted pose of the distance sensor from the actual pose and / or in the surface model from the actual working area. The additional virtual measurement thus represents another conceivable prediction of the actual measurement. A comparison is then made to determine whether the actual measurement is compatible with the virtual measurements, i.e., with the first virtual measurement and the additional virtual measurement, or with the additional virtual measurements themselves.Compatible means, in particular, that the actual measured value agrees with at least one of the virtual measured values, whereby, instead of theoretical identity, a tolerance may be permitted, for example, in the form of a threshold, an absolute or relative positional error. If the actual measured value is incompatible with the virtual measured values, an object is inferred from the distance measuring beam. The actual measured value then does not fit the surface model and is therefore not explainable by the known contour or the objects known to be present in the working area.

[0017] This is a computer-implemented method that runs, for example, in the distance sensor, a safety controller, or other computing unit connected to the at least one distance sensor and the machine. Implementation in the machine control system or a distributed implementation is also conceivable. Furthermore, it should be noted that the generation of virtual distance measurement beams merely illustrates the corresponding calculations, and it is not necessary for a beam to be actually represented in the implementation. In many cases, a much simpler calculation can be performed, which ultimately determines the point of impact of virtual distance measurement beams.

[0018] The invention has the advantage of providing a particularly robust method for object detection. By incorporating at least one additional virtual distance measuring beam, further information about the expected actual measured values ​​is obtained. This allows for a significantly improved ability to judge whether an object is actually located within the distance measuring beam. False detections caused by deviating virtual measured values, especially at edges of the surface model, are thereby prevented. The invention facilitates the possibilities discussed in the introduction for ensuring the safety of movements near a work surface or of dynamic movements, or even makes them accessible to practical application in the first place.

[0019] Preferably, for a detected object, a check is performed to determine whether its presence is permissible; otherwise, the machine is switched to a risk-reducing or safe state. A detected object does not necessarily mean that the machine requires safeguarding. Depending on the safety concept, an object may be tolerated for short periods, ignored if it is too small a source of interference, located sufficiently far from the actual hazard and, in particular, outside of protective zones, and so on. However, if a detected object does not fall under any of these defined permissibility conditions, a signal is triggered that switches the machine to a safe state. Risk-reducing means that at least serious consequences of accidents are avoided, whereby accidents are prevented by a safe state within the scope of the achieved safety level.Depending on the application and situation, this can be achieved through measures such as slower movements, a restricted working area, or, if necessary, stopping the machine.

[0020] The surface model is preferably configured or imported, especially as a CAD model, or the surface model is measured with a sensor. Measuring the topography is not strictly necessary; a 3D model can also serve as the basis for the acquisition, such as a CAD model, which is often readily available for many machines and also in the case of robotic cells. If measurement is used, acquisition without disruptive machine movements is advantageous. In principle, the distance sensor itself can be used for the measurement. A dedicated topography sensor, such as a laser scanner or a camera, especially a 3D camera, is also conceivable. The topography sensor can also be very expensive or complex and require specialized knowledge to operate, as it is only needed for a short time. The topography sensor can move with the machine or be mounted externally.Configuration, import, or measurement of the surface model is preferably performed in advance. Alternatively or additionally, the surface model can be at least partially reloaded during a production interruption or even dynamically at runtime. This is useful, for example, when a tray is replaced with another because the production process is switching to a different part. This production changeover can trigger the reloading of the corresponding surface model automatically or manually. A modular surface model facilitates such changes; only the modified modules or sections then need to be dynamically replaced. Furthermore, in the case of manual modifications to a CAD model, it is advantageous to be able to adjust the surface model used with a simple command, such as a mouse click, after completing the manual changes.

[0021] The first virtual distance measuring beam is preferably surrounded by a multitude of further virtual distance measuring beams. This covers an entire range of possible deviations between the first virtual distance measuring beam and the real distance measuring beam of the distance sensor. Object detection becomes particularly robust.

[0022] The starting points of the subsequent virtual distance measurement rays preferably form, at least topologically, a circle. These subsequent virtual distance measurement rays thus cover deviations in all conceivable directions. Effectively, they form a kind of surface extending from the circle of their starting points or their respective origins. The first virtual distance measurement ray is embedded within this surface. A circle, at least topologically, can be distorted and thus, for example, elliptical or otherwise warped, including the possibility of having corners like a square. However, a circle in the geometric, and not just topological, sense is particularly preferred because it is the simplest to handle and best represents the surroundings.

[0023] The surface model is preferably a polygon mesh, in particular a triangular mesh. A polygon mesh allows the contour of the work environment to be quantized in a suitable data structure with any desired precision. The polygon mesh is composed of polygon surface elements, which are referred to as polygons in the following.

[0024] The additional virtual distance measurement beams preferentially intersect polygons that are seamlessly adjacent to each other. In other words, the polygons that the virtual distance measurement beams intersect are connected. This essentially describes the density of the additional virtual distance measurement beams, which, within the accuracy limits of the polygon mesh, is intended to ensure seamless object detection. The density condition can be calibrated against the largest conceivable distance in the surface model, because for smaller distances, the polygons intersected by the additional virtual distance measurement beams are then even more seamlessly adjacent. The condition of seamlessly adjacent polygons does not have to be universally applicable; for example, it can be limited to surfaces at a specific angle of inclination to the sensor.For example, if the distance measurement beams graze the target at a very shallow angle, an infinite density of distance measurement beams would otherwise have to be generated in the limit; such cases are not of practical interest. It is particularly preferred that, at least at the greatest conceivable distance, each of the seamlessly adjacent polygons is hit exactly once. Multiple hits by several of the additional virtual distance measurement beams do not compromise safety but waste computing time without providing new information. In an embodiment where the starting points of the additional virtual distance measurement beams form a circle, the density condition can be formulated using the angular spacing or circumferential spacing of the additional virtual distance measurement beams, which must be chosen to be correspondingly small.The seamlessly adjacent polygons then form a circular arc within the limits of their discrete resolution, which may be distorted in perspective to an elliptical arc and may be divided by edges in the working environment.

[0025] Preferably, virtual distance measurements are determined for polygons enclosed by polygons onto which further virtual distance measurement beams intersect. This effectively completes or closes off the portion of the surface model considered by these further virtual distance measurement beams. In an embodiment where the starting points of the further virtual distance measurement beams form a circle, at least topologically, the entire circular area is considered, and not just the circumference. As mentioned above, the intersecting virtual distance measurement beams are primarily an illustration and do not necessarily reflect the actual implementation. Enclosed polygons and their associated virtual measurements can be determined much more easily and with significantly less effort by calculating the distance between two points in space.

[0026] The starting points of the additional virtual distance measurement beams preferably form a grid. This is an alternative way to consider a planar neighborhood of the surface model instead of just a line, in particular a surrounding line in the form of at least a topological circle. In many implementations, it will be less complex to choose the embodiment presented in the preceding paragraph, which requires fewer additional virtual distance measurement beams for the same considered area of ​​the surface model. The density conditions mentioned above are preferably applied to the mutual distances of the starting points so that all polygons within the grid area are intersected by at least one additional virtual distance measurement beam. The grid is preferably regular, since it is difficult to generally satisfy the density condition with irregular spacing, which is conceivable in principle.It should be reiterated that it is not necessary to form a grid of virtual distance measurement beams in the implementation in the sense of a direct representation; any calculations, some of which are considerably simpler, are sufficient to obtain the corresponding virtual measurement values.

[0027] The subsequent virtual distance measuring beams preferably run parallel to the first virtual distance measuring beam. This also means that the subsequent virtual distance measuring beams preferably run parallel to each other. In an embodiment where the starting points of the subsequent virtual distance measuring beams lie, at least topologically, on a circle, the subsequent virtual distance measuring beams accordingly form a cylindrical shell. In principle, obliquely running subsequent virtual distance measuring beams are conceivable, particularly with a conical shell instead of a cylindrical shell, but this introduces avoidable distance dependencies. In safety engineering, it is generally not advantageous if only larger objects are reliably detected at greater distances; rather, a detection capability for the entire working range is defined.In some designs, an acute angle, at which the distance measuring beams converge, can still be useful.

[0028] The actual measured value and the virtual measured values ​​are preferably recalculated and compared cyclically. For each measurement cycle, ideally for the highest time resolution, the virtual measured values ​​are updated. This results in a detection cycle for object recognition.

[0029] Preferably, an actual measurement that is larger than all virtual measurements is considered acceptable, and thus, in this case, no object is detected in the distance measurement beam. In this situation, the actual measurement does not match any virtual measurement within the tolerances. However, the object is located further away than the working area, and this is safe from a safety perspective. Handling an excessively large actual measurement in this way allows the surface model to be simplified, for example, by omitting dents (i.e., outwardly curved partial contours of the working environment from the perspective of the distance sensor) and replacing them with a flat surface. Alternatively, according to the quiescent current principle, a measurement that confirms the model is always required, meaning that even an excessively large actual measurement is not tolerated.

[0030] The distance sensor preferably has several distance measuring beams, with which a respective actual measurement value for the distance is determined and compared with virtual measurement values ​​from virtual distance measuring beams for a respective distance measuring beam. In this embodiment, there are several real measurements, wherein the corresponding distance measuring beams originate from the same distance sensor, each distance measuring beam originates from one of several distance sensors, or single- and multi-beam distance sensors are used in combination. The above procedure is carried out for each of these distance measuring beams, i.e.,The respective distance measuring beam is simulated by a virtual distance measuring beam, and at least one further virtual distance measuring beam offset from it is formed in order to detect an object in the considered distance measuring beam by comparing the actual measured value of the considered distance measuring beam with the virtual measured values ​​of the associated virtual distance measuring beams.

[0031] The distance measuring beams preferably form a protective dome, whereby no object is detected within the distance measuring beam if the actual measured value exceeds the length of the protective dome. The protective dome is positioned around the tool tip on a robot arm, enabling timely detection of approaches to the tool tip. The protective dome does not necessarily have to be completely enclosed, for example, if access from the rear is not possible for mechanical or other reasons. Objects at a distance greater than the length of the protective dome are not relevant from a safety perspective and are therefore ignored in object detection in this embodiment. As the robot approaches the work area, the protective dome can be adapted to the surface model, as access to the shortened areas of the protective dome is not possible but is blocked by sections of the work area.In this embodiment, particular care must be taken not to confuse the distance measuring beams and the virtual distance measuring beams. The distance measuring beams themselves already form a cylindrical or conical shell. In the preferred embodiment, with the starting points of further virtual distance measuring beams on a circular line, a separate virtual cylindrical or conical shell is formed around each distance measuring beam.

[0032] The distance measuring beam preferably performs a scanning motion. This is an alternative to using multiple distance measuring beams to achieve a larger object detection area. Both approaches can be combined, as in the case of a multi-beam laser scanner. The virtual distance measuring beams are generated for the respective scan position of the distance measuring beam.

[0033] The non-contact distance sensor according to the invention for safeguarding a machine, in particular a robot arm, and for detecting objects in a working area of ​​the machine, comprises a control and evaluation unit configured to measure an actual measured value for a distance to the working area or an object located therein and to detect an object in the distance measuring beam based on the actual measured value using a method according to the invention. It is, in particular, an optoelectronic distance sensor based on the time-of-flight principle, wherein the distance sensor has a light transmitter for emitting a light signal along a distance measuring beam and a light receiver for generating a received signal from the returning light signal, and the control and evaluation unit determines the actual measured value from the time of flight of light between the emission and reception of the light signal.

[0034] The control and evaluation unit is preferably not integrated exclusively into the distance sensor itself, but at least partially outsourced to a connected system, such as a safety controller, a non-safety controller, or another computing unit, which is then secured through redundancy or testing. The machine control system can also incorporate at least parts of the control and evaluation unit. Implementing the unit outside the distance sensor facilitates updates and adaptation to other applications, and also allows for a simpler design of the distance sensor. However, a control and evaluation unit integrated into the distance sensor, as well as a distributed implementation across the distance sensor and connected computing units, are also conceivable.

[0035] The distance sensor is preferably mounted on the end effector of a robot and protects it with a protective bell formed by several distance measuring beams (EOAS, End of Arm Safeguarding). Mounting on another moving machine is also conceivable, such as an AGV (automated guided vehicle), particularly in an embodiment as a laser scanner that performs a scanning motion with the distance measuring beam. This scanning motion can also be merely emulated in a solid-state lidar. In another possible embodiment as a 3D camera, particularly a TOF camera (time-of-flight, 3D camera with measurement of the time of flight in its pixels), each pixel spans a distance measuring beam, whereby pixels can be grouped or selected to produce specific distance measuring beams.

[0036] The distance sensor is preferably a safety sensor and / or its functionality is cyclically checked and / or the distance values ​​of several distance sensors are compared to generate safe distance values. Thus, safe distance values ​​are obtained. The distance sensor itself can already be designed in accordance with standards. Safety can also be achieved through multiple redundant and / or diverse redundant distance sensors or cyclical functional tests, or a higher level of safety can be achieved in this way. Terms such as "safe" or "safety sensor" are always to be understood in this context as meaning that a safety standard for applications in safety engineering or for accident prevention in the industrial sector, in particular for machine safety, non-contact protective devices, industrial robots, collaboration with robots, or the like, is met.These can be the norms mentioned in the introduction or their successors, extensions or respective equivalents in other regions of the world.

[0037] The invention is further explained below with regard to additional features and advantages by way of example embodiments and with reference to the accompanying drawing. The illustrations in the drawing show: Fig. 1. An overview of a machine to be secured, using the example of a robot arm with a distance sensor attached to it, which creates a virtual protective bell; Fig. 2 a block diagram of a safety system with at least one distance sensor; Fig. 3 a schematic representation of a distance measuring beam and virtual distance measuring beams; Fig. 4 a representation of a polygon mesh as a surface model of a work environment, with illustration of the points of impact of a distance measuring beam and virtual distance measuring beams; Fig. 5 a schematic representation of a distance measuring beam and virtual distance measuring beams as well as the virtual measured values ​​determined therewith when measuring at an edge; Fig. 6 a representation according to Fig. 5 now when measuring on a flat surface; Fig. 7 a schematic representation to illustrate a simplified surface model when excessively large actual measured values ​​are tolerated; and Fig. 8 A schematic representation of a measurement situation in which an object is not detected on an inclined surface facing the object, to explain why this is not a safety-critical issue.

[0038] Fig. Figure 1 shows an overview of a robot arm 10 to be secured. The robot arm 10 and the specific application are examples, and the following explanations are transferable to any robot and other moving machines and scenarios to be secured, in particular AGVs / AGCs (Automated Guided Vehicle / Container) or drones.

[0039] To safeguard the end effector, specifically at its tip, 10 distance sensors 12a-b are mounted on the robot arm, preferably in the vicinity of a tool for end-of-arm safeguarding (EOAS). The distance sensors 12a-b determine distance values ​​along several distance measurement beams 14. The number of two distance sensors 12a-b shown is purely exemplary; there can be more distance sensors or just one, which can then preferably measure along several distance measurement beams 14. In general, one or more distance measurement beams 14 emanate from each distance sensor 12a-b. Distance measurement beams 14 can be approximately geometric beams or have a finite cross-section, for example, if the distance sensor 12a-b operates as an area sensor with a fanned-out light beam. Optoelectronic distance sensors, for example, with time-of-flight (ToF) measurement, are particularly suitable as distance sensors 12a-b.The aforementioned DE 10 2015 112 656 A1 presents such a system, to which reference is made for further information. However, other optoelectronic sensors exist for determining distances, such as laser scanners and 2D or 3D cameras, as well as entirely different technologies, such as ultrasonic sensors, capacitive sensors, radar sensors, and the like.

[0040] The distance measuring beams 14 together form a kind of virtual protective shell or virtual protective dome 16 around the end effector. If a person, for example, reaches into the area protected by the protective dome 16 with their hand 18, the protective dome 16 is considered violated. Therefore, a safety-related reaction of the robot 10 is triggered, which, depending on the violated distance thresholds, can consist of slowing down, evading, or an emergency stop.

[0041] The detection of the hand 18 in a distance measuring beam 14 is carried out by means of object recognition, which is described below with reference to the Fig. Sections 3 to 8 are explained in more detail. In brief, a surface model of the work area 20, including the objects 22 expected on it, is known. The surface model corresponds to the 3D contour or topography of the work area 20 with the objects 22 expected on it and is preferably known in advance. For example, it is created manually in a CAD model or an existing surface model, such as that of the robot cell, is imported. Furthermore, a contour measurement can be performed, either with the distance sensor 12a-b itself or with a topography sensor, such as a laser scanner or a 3D camera, attached externally and / or to the robot 10 for teaching the surface model. Alternatively or additionally, instead of creating the surface model in advance, the surface model can also be created or modified during operation or during an interruption of operation.

[0042] Based on the surface model, the pose of the distance sensor 12a-b can be used to predict the distance that should be measured, provided no object, and in particular no hand 18, is in the distance measuring beam 14. The pose, in turn, is calculated from the robot's movements, this so-called forward kinematics, which can generally be determined and queried for robot control anyway. The measuring behavior of the distance sensor 12a-b is thus simulated, generating a virtual measurement value that can be compared with the actual measurement value of the distance sensor 12a-b to reveal whether something other than what is expected according to the surface model is in the distance measuring beam 14. It should be noted that expected objects 22 are part of the surface model and are therefore ignored by object recognition.To make the comparison robust, according to the invention not only is the distance measuring beam 14 simulated, but at least one additional virtual measured value is determined and compared from at least one further virtual distance measuring beam adjacent to the simulated distance measuring beam. This increases the robustness of the method, particularly in the case of edge hits or near depth changes in the 3D contour of the work surface 20 with expected objects 22 located thereon or the corresponding surface model.

[0043] As in Fig. As can be seen in Figure 1, a limited length of the protective bell 16 is sufficient for safety. Therefore, measured values ​​from the distance sensor 12a-b greater than this length can preferably be ignored in object detection, since it is not safety-relevant whether an object is located at such extreme distances or not.

[0044] Fig. Figure 2 shows a block diagram of a safety system according to the invention, wherein the robot arm, as already mentioned as a possibility, is now generalized to a machine 10. One or more distance sensors 12 are arranged as in Fig. The distance sensors shown in Figure 1 move along with the machine part to be protected, in order to safeguard it against interference and thus prevent accidents and injuries. The distance sensors 12 are preferably optoelectronic sensors based on the time-of-flight principle, but other similar or mixed sensors from the examples mentioned above are also conceivable.

[0045] The distance values ​​are transmitted to a control and evaluation unit 24, where the object detection, which is yet to be described, is performed and, if necessary, a safe signal is output to the machine 10 or its controller to trigger a safety response. The control and evaluation unit 24 can be, as shown, an external computing unit, alternatively an internal computing unit of the at least one distance sensor 12, or a combination of both. Examples of an internal computing unit are digital processing components such as a microprocessor or a CPU (Central Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), an ASIC (Application-Specific Integrated Circuit), a K-processor, an NPU (Neural Processing Unit), a GPU (Graphics Processing Unit), or the like.An external computing unit can be any type of computer, including notebooks, smartphones, tablets, (security) controllers, as well as a local network, an edge device, or a cloud. There is also a wide selection of communication connections, such as I / O Link, Bluetooth, WLAN, Wi-Fi, 3G / 4G / 5G, and essentially any industry-standard protocol.

[0046] The distance sensors 12 and the control and evaluation unit 24 are preferably intended to provide safe data acquisition and evaluation in accordance with the relevant standards. One way to achieve this is by using safe sensors and a safety controller. However, it is also conceivable to use inherently non-safe hardware and achieve the required level of safety through redundancy, including diverse redundancy, or through plausibility checks or cyclical tests. The object recognition process now described contributes to fulfilling the safety requirements.

[0047] Fig. Figure 3 shows a schematic representation of a distance measuring beam 14 running close to an edge 26. The distance sensor 12 measures the distance to the point of impact of the distance measuring beam 14, which, depending on the situation, lies on the work surface 20, an object 22 expected there, or an object to be detected, such as a hand 18. In contrast to the following simulations, the measured distance is referred to as the actual measured value. In the following, it is assumed that distance measuring beams 14 are considered individually. The procedure is analogous if several distance measuring beams 14 are combined and, for example, their mean value is used as the corresponding actual measured value. Furthermore, no knowledge of the path of motion of the distance sensor 12, which moves with the machine 10, is assumed.Alternatively, a reference run could simply be carried out beforehand, in which the expected measured values ​​for the situation without detectable objects, including expected errors, are taught in.

[0048] Since the current pose of the distance sensor 12 is unknown a priori, it is first determined. For this purpose, the forward kinematics of a robot controller or comparable data from another machine controller are used. Any pose offset between the machine and the distance sensor 12 can be compensated for by initial configuration or calibration. From this pose, the assumed path of the distance measuring beam 14 is then calculated. This is an assumed or simulated distance measuring beam as a virtual equivalent of the distance measuring beam 14, which is referred to here as the first virtual distance measuring beam 28. Using the first virtual distance measuring beam 28, a first virtual measurement value for the distance is calculated by determining the point of impact based on the surface model.A hand 18 in the distance measuring beam 14 would significantly shorten the actual measured value, so that a comparison between the actual measured value and the first virtual measured value detects an object in the distance measuring beam 14.

[0049] However, as is also the case in Fig. As shown in Figure 3, the distance measuring beam 14 and the first virtual distance measuring beam 28 do not exactly coincide due to inaccuracies, particularly in pose determination and / or the surface model. At edge 26, this may lead to significant differences between the actual measured value and the first virtual measured value, and thus to false-positive object detections.

[0050] Therefore, according to the invention, at least one further virtual measurement value is generated by placing at least one further virtual distance measurement beam 30 alongside the first virtual distance measurement beam 28 and calculating the distance to its point of impact based on the surface model. The further virtual distance measurement beam 30 has an offset 32 ​​to the first virtual distance measurement beam 28 and preferably runs parallel to it. No object is considered detected as long as the actual measurement value is compatible with the virtual measurement values, i.e., the first virtual measurement value and the at least one further virtual measurement value. For this purpose, for example, a comparison of the actual measurement value with the virtual measurement values ​​is carried out, and no object is considered detected as long as at least one of these comparisons yields a match, preferably allowing for a tolerance.This covers both situations: the distance measuring beam 14 hitting the object 22 or the work surface 20 on either side of the edge 26 or on the other side. The limited offset 32 ​​ensures that the special situation of very different virtual measured values ​​remains confined to the vicinity of the edge. This prevents any detectable object with dimensions larger than the offset 32 ​​from being overlooked.

[0051] With only one additional virtual distance measuring beam 30, conceivable deviations of the simulation are only taken into account in a single direction. Preferably, therefore, a plurality of additional virtual distance measuring beams 30 are generated, which are handled accordingly. Preferably, an additional virtual distance measuring beam 30 is provided in each direction, so that the starting points of the additional virtual distance measuring beams 30 lie on at least a topological, preferably also a geometric, circle. The difference between topological and geometric means, as is common in mathematics, that only in the former case are deformations permitted; thus, for example, an ellipse or a square is also a topological circle. The case of a geometric circle is explained in more detail below as an example, without limiting the invention to this case.

[0052] Fig. Figure 4 shows a representation of a polygon mesh as a preferred representation of the surface model. This representation is advantageous, but merely exemplary. Triangle polygons (Triangle Mesh) are particularly preferred. The representation is a top view of a neighborhood around edge 26. To the right of edge 26, there is an area shadowed by object 22 of the surface model. Geometrically, no rays emanating from distance sensor 12 can strike this area.

[0053] Due to the aforementioned discrepancies between reality and simulation, the distance measuring beam 14 and the first virtual distance measuring beam 28 encounter different polygons, deviating from the representation of the Fig. 3 now on this side and beyond edge 26. The further virtual distance measuring beams 30 describe, according to the assumed embodiment, circular or, depending on the inclination of the polygons, more generally elliptical curve segments 33 with starting points on a circle on the polygon grid. Further virtual measured values ​​are taken for the areas swept out by the curve segments 33, in Fig. Four dark-colored polygons 34 are calculated. Additionally, further virtual measurements are also calculated for the inner polygons 36, which are surrounded by the curve segments 33 and shown in lighter colors. Thus, ultimately, in reverse order, another virtual distance measurement beam 30 is evaluated for each polygon of the area enclosed by the curve segments 33. In the example shown, the first virtual measurement of the first virtual distance measurement beam 28 coincides with one of the inner polygons 36. This redundancy is harmless but can also be handled in the implementation.

[0054] The virtual measured values ​​can be stored in an array and updated cyclically, in particular with each measurement cycle of the distance sensor 12. In corresponding evaluation cycles of the object detection, again preferably with each measurement cycle of the distance sensor, a comparison is made to determine whether the actual measured value is compatible with the virtual measured values. As already explained, this comparison specifically checks whether there is at least one virtual measured value that is equal to the actual measured value within tolerances. If this is the case, the actual measured value can be explained by the surface model, and no object or hand 18 has been detected. Otherwise, an object must be present and have interrupted the distance measuring beam 14, because otherwise the actual measured value would not deviate from all virtual measured values ​​by more than the tolerance.

[0055] To ensure that the polygons 34 under the curve segments 33 are connected, i.e., without gaps, the additional virtual distance measurement beams 30 are preferably positioned at a correspondingly close distance. This can be calibrated on a surface at the maximum possible distance from the surface model. The arrangement discussed here as an example, with the starting points of the additional virtual distance measurement beams 30 on a circular line, can be understood as the rotation of another virtual distance measurement beam 30 on a circle with a radius o corresponding to the offset 32 ​​and a certain angular increment w. The offset 32, and thus the radius o, is preferably chosen such that, for all expected errors, the distance measurement beam 14 is located within the cylindrical shell spanned by the additional virtual distance measurement beams 30 in the position of the distance sensor 12 or the surface model.Furthermore, preferably the angle increment w is so small that the above density condition is satisfied, i.e. the polygons 34 are connected under the curve segments 33.

[0056] Fig. Figure 5 shows the schematic representation of the diagram on the left. Fig. Figure 3 illustrates the virtual measurements in the vicinity of edge 26 on the right. The spread between the minimum and maximum virtual measurements is large due to edge 26. Intermediate values ​​are not represented by virtual measurements; the transition at edge 26 is abrupt. An actual measurement that is compatible with the surface model, and thus does not indicate an object, must lie within the tolerance ranges 38 around the virtual measurements. If an actual measurement lies within the large remaining sensitive range 40, then an object is detected.

[0057] Fig. Figure 6 shows a representation according to Fig. 5 now without edge 26 when measuring on a flat surface. Here, the actual measured value is only compatible with a very small range of distance values ​​from the outset, since the minimum and maximum virtual measured values ​​are very close to each other, resulting effectively in almost only a single tolerance range 38. Again, with an actual measured value in the remaining sensitive range 40, an object is detected.

[0058] Fig. Figure 7 shows a schematic representation of a surface model, where the contour of the work surface 20 and the expected objects 22 located on it are combined. In this example, there are small, locally confined depressions 42, such as those frequently found on a robot's work surface 20, for example, in trays. In a preferred embodiment of the invention, a simplified surface model 43 is used that ignores these depressions 42 and instead assumes continuous surfaces. This simplification leads to the actual measured value being too large in the area of ​​a depression 42. Accordingly, object detection is only triggered by an actual measured value that is too small, but not by one that is too large. In particular, for this special case, the actual measured value must be larger than all virtual measured values ​​combined.This is permissible from a safety perspective, because it is still impossible that an object has come dangerously close to machine 10.

[0059] Fig. Figure 8 shows a schematic representation of a special measurement situation in which an exemplary object 44, which should normally be detected by object recognition, is overlooked, although this is not safety-critical. This object 44 is located on a surface inclined towards the distance sensor 12. More generally, there is a depth discontinuity at object 22 of the surface model or its edge, but the other virtual distance measurement rays 30 also intersect the connecting surface, which is perpendicular here, so that intermediate values ​​are recorded as virtual measured values. As shown on the right in Fig. As illustrated in Figure 8, the intermediate values ​​with their respective tolerance ranges 38 complement each other to form a continuum in an intermediate range between the minimum and maximum virtual measured value. The sensitive range 40 of the object detection is significantly reduced and excludes this intermediate range. Therefore, an actual measured value for object 44 is compatible with the virtual measured values, and the object detection does not trigger on object 44. This is the situation of Fig. 5 is completely different, where this intermediate area lies outside the tolerance ranges 38 and is a sensitive area 40.

[0060] Although object recognition is therefore incomplete, this is irrelevant for security purposes. It is physically impossible for hand 18 to reach the location of object 44 without first being detected in the protective dome 16, which is located in Fig.The hazard 46 represented by a gripper is secured. Any other object 48 that might attempt to reach the position of object 44 is indeed located within the sensitive area 40.

[0061] As an alternative to the described procedure, it is conceivable to establish a highly accurate match between distance measuring beam 14 and the first virtual distance measuring beam 28 through calibration, thereby generating a very precise surface model and calibrating it against the distance sensor 12 and the real environment. However, this is complex and not necessarily stable over the long term, so that false-positive object detections may occur again after a certain period of operation. Nevertheless, a one-time or repeated calibration is conceivable as a supplementary measure. In particular, after calibration, the offset 32 ​​or the radius o introduced above can remain small, while still ensuring that the distance measuring beam 14 lies within the other virtual distance measuring beams 30, thus limiting the effort required for evaluation.

Claims

[1] Method for securing a machine (10), in particular a robot arm, in which objects (18) in a working area (20, 22) of the machine (10) are detected by at least one contactless distance sensor (12) moving with the machine (10) measures an actual measured value for a distance to the working area (20, 22) or an object (18) arranged there, with at least one distance measuring beam (14), from a pose of the distance sensor (12), a first virtual distance measuring beam (28) is formed, which simulates the distance measuring beam (14), the point of impact of the first virtual distance measuring beam (28) on a surface model of the work area (20, 22) is determined and a first virtual measured value for the distance is calculated and the actual measured value is compared with the first virtual measured value to decide whether an object (18) is located in the distance measuring beam (14), characterized by , that at least one further virtual distance measuring beam (30) is formed with an offset (32) to the first virtual distance measuring beam (28), the point of impact of the at least one further virtual distance measuring beam (30) on the surface model is determined and at least one further virtual measured value for the distance is calculated, that a comparison is made as to whether the actual measured value is compatible with the virtual measured values, in particular whether it is equal to at least one of the virtual measured values ​​up to a tolerance (38), and that an object (18) is detected in the distance measuring beam (14) if the actual measured value is not compatible with the virtual measured values. [2] Method according to claim 1, wherein for a detected object (18) it is checked whether its presence is permissible and otherwise the machine (10) is transferred to a risk-reducing or safe state. [3] Method according to claim 1 or 2, wherein the surface model is configured or imported, in particular as a CAD model, or wherein the surface model is measured with a sensor. [4] Method according to one of the preceding claims, wherein the first virtual distance measuring beam (28) is surrounded by a plurality of further virtual distance measuring beams (30), wherein in particular the starting points of the further virtual distance measuring beams (30) form at least topologically a circular line. [5] Method according to one of the preceding claims, wherein the surface model is a polygon mesh, in particular a triangular mesh. [6] Method according to claim 5, wherein the further virtual distance measuring beams (30) strike polygons (34) which are adjacent to one another without gaps. [7] Method according to claim 5 or 6, wherein virtual measured values ​​for the distance are determined for polygons (36) which are enclosed by polygons (34) on which further virtual distance measuring beams (30) impinge. [8] Method according to one of the preceding claims, wherein the starting points of the further virtual distance measuring beams (30) form a grid. [9] Method according to one of the preceding claims, wherein the further virtual distance measuring beams (30) run parallel or at an acute angle to the first virtual distance measuring beam (28). [10] Method according to one of the preceding claims, wherein the actual measured value and the virtual measured values ​​are cyclically re-determined and compared. [11] Method according to one of the preceding claims, wherein an actual measured value which is greater than all virtual measured values ​​is assessed as compatible and thus no object (18) is detected in the distance measuring beam (14) in this case. [12] Method according to one of the preceding claims, wherein the distance sensor (12) has a plurality of distance measuring beams (14) with which a respective actual measured value for the distance is determined and compared with virtual measured values ​​from virtual distance measuring beams (28, 30) for a respective distance measuring beam (14). [13] Method according to one of the preceding claims, wherein the distance measuring beams (14) form a protective bell (16) and wherein in particular no object (18) is detected in the distance measuring beam (14) if the actual measured value exceeds a length of the protective bell (16). [14] Method according to one of the preceding claims, wherein the distance measuring beam (14) performs a scanning movement. [15] Contactless distance sensor (12), in particular an optoelectronic distance sensor based on the time-of-flight principle, for securing a machine (10), in particular a robot arm, and for detecting objects (18) in a working area (20, 22) of the machine (10), wherein the distance sensor (12) has a light transmitter for emitting a light signal along a distance measuring beam (14), a light receiver for generating a received signal from the returning light signal and a control and evaluation unit (24) which is designed to measure an actual measured value for a distance to the working area (20, 22) or an object (18) arranged there and to detect the object (18) in the distance measuring beam (14) on the basis of the actual measured value using a method according to one of the preceding claims.

Images