Sensor-robot tandem boosts solar-cell yield

A leading manufacturer of solar-panel cells has modified an existing materials-handling and machine-vision system that limited the products' reliability and production volume.

Jan 1st, 2002
Th 82463

By Michael Williams

Smart sensor works with a robot to rapidly and accurately position, image, and place solar cells for increased production and fast fault determination.

A leading manufacturer of solar-panel cells has modified an existing materials-handling and machine-vision system that limited the products' reliability and production volume. Within a month after installing new vision and imaging hardware and software, the manufacturer experienced a record production level, most of which the company attributes to the improved performance of the upgraded vision inspection system.

The original inspection system was slow and failed to detect many nonconforming parts. Furthermore, it was hardware-based, affording the company no fast and easy way to make modifications. The manufacturer also used outdated software, which restricted substantial improvements in the inspection process. The new system has yielded a 15% increase in production volumes and can identify and remove all nonconforming parts.

Click here to enlarge image

FIGURE 1. During operation of vision-inspection system, robot arm removes a solar-panel cell from the stack and positions it over the vision window. After cell placement, the sensor (camera), located beneath the window, exposes images of the cell. Strobe lights near the camera are synchronized to fire when exposures are made, and LED modules above the vision window (not shown) provide contrast for better imaging. The images are checked for defects by software before the robot places the cell to the drop-off position.

Boston Engineering Corp. (Waltham, MA) was selected to develop and install the new vision inspection system, partly because the company guaranteed delivery and installation in 4.5 months. To assist with the project, Boston Engineering contacted Gibson Engineering Co. Inc. (Norwood, MA), which is a DVT-certified automation solution provider.

Boston Engineering offers systems-integration platforms to companies requiring machine-vision design and control-development assistance. The company has imaging experience with products from Cognex Corp. (Natick, MA), DVT Corp. (Norcross, GA), Opteon Corp. (Cambridge, MA), and RVSI Acuity CiMatrix (Canton, MA). The vision inspection system developed for the solar-panel cell manufacturer required the use of a selectively compliant articulated robot arm (SCARA) robot working in tandem with a DVT Corp. Series 600 SmartImage Sensor (camera).

Mark Smithers, vice president of business development and chief operating officer of Boston Engineering, says, "We met a demanding development schedule for a critical operation. Had the system not been delivered on time or not worked well, the client would have suffered significant revenue losses."

The new vision inspection and handling system was installed during a one-week shutdown at the manufacturer's site. The system was set up to be a slave to the existing system, which is based on a programmable logic controller (PLC) from Allen-Bradley Inc./Rockwell Automation (Milwaukee, WI). Because there were many unused signals from the PLC, it was also necessary to "clean up" the wiring and PLC software by simplifying the interface to the robotic positioner and removing unnecessary code. Design challenges that were confronted included removal of some old equipment, replacing it with new equipment, and avoiding impacting production levels, all in 4.5 months; adding a human-machine interface (HMI) and a new conveyor subsystem after the project started; redocumenting the existing system and changing the PLC code without affecting the existing functions to be retained; and integrating a robot-vision parts-handling system.

The old equipment used pushbuttons and lamps to indicate operational conditions. The new HMI allows operators to control or modify parts placement to address upstream variances. Instead of throwing away defective parts, the new handling equipment can be adjusted to accommodate them. For master PLC programming, the inspection operations are documented so that other developers who need to modify them have an easier time understanding what had been done and what the code is instructing the machine to do.

Dumont Associates Inc. (Nashua, NH), a controls service provider and Boston Engineering's development partner, was contracted to handle the PLC programming and documentation.

The key components of the new vision system are the SCARA robot and a Super SEL-type IH controller, both supplied by Intelligent Actuator Inc. (Torrance, CA); a DVT Series 600 SmartImage Sensor; a conveyor; a Mitsubishi Electric Automation (Vernon Hills, IL) E series HMI; safety interlock doors; high-intensity strobe lighting and LED illumination; and protective light curtains. These components were designed to work as an integrated module with seamless communications. The sensor was configured to transmit signals to the robot (that is, robot-executed commands) and to serve as the visual data source. Gibson Engineering is the manufacturers' representative for the DVT, Intelligent Actuator, and Mitsubishi products in the module.

FIGURE 2. Red-emitting LED modules mounted at left and right above the vision window provide contrast to the solar-panel cells being inspected. A circular hood (white disk above the LED modules) reflects the light downward onto the solar cell being inspected?a technique that improves image contrast while keeping the area clear for the robotic arm.
Click here to enlarge image

Communications consist of an RS-422 serial link from the DVT sensor to a converter that changes the signals to RS-232 signals and then delivers them to the robot. The robot, in turn, sends the RS-232 signals to the HMI. An Ethernet connection to the robot receives system operational data through a remote terminal, either while running in production mode or in manual debug mode. The Ethernet connection serves as a tool for analysis and debug and a means for adding system features.

Most of the programming is performed in the sensor using a laptop computer running DVT Windows-based point-and-click software—FrameWork 2.3. The IH-600 SCARA robot was selected because it offers four axes of motion (x, y, z, and rotation at the end effector), 1-g maximum acceleration, and 4-m/s maximum speed, all at a reasonable price.

The HMI—the Mitsubishi model E300—communicates to the robot controller through a serial RS-232 interface. It provides several password-protected windows that allow the operator to view all the system inputs, including the PLC commands and the sensor inputs. Using the HMI, the operator also can change the type of solar-panel cell inspection to be performed. Another feature is an error log, which takes data from the sensor and logs/time-stamps the type of failure of rejected cells.

Using FrameWork software, Boston Engineering was able to "teach" the sensor to look for various features to ensure compliance with "go/no-go" features. Some additional custom code was developed for cell-transformation calculations and communications to the robot.

Smithers notes that Boston Engineering was able to take advantage of the coordinate-system soft-sensor feature incorporated into the SmartImage sensor that allows users to specify motion-control x and y coordinates. FrameWork then transforms the output parameters from the sensor's pixels to the user's set of points. "This tool is indispensable for those who are integrating motion-control systems with a sensor," Smithers says.

Most users apply blob-tool soft sensors to calculate image coordinates for objects and then apply the coordinate system soft sensor to convert those points into real-world values for the robot. Smithers says, "The DVT coordinate system soft sensor allowed us to map the vision space with the robot space, eliminating one coordinate-transform function and dramatically streamlining the positioning algorithm. It proved to be an important contributor to getting the system up and running under a short development schedule," adds Smithers.

In operation, the Boston Engineering platform picks a solar-panel cell from a tray of cells, presents it to the vision inspection system, and then places the inspected cell at a predetermined position for the next process. The system accuracy requirements are demanding. Parts placement has to be ±0.01 in. in the x and y axes for a selected point on the cell, has to be correct for a ±9-mm source (tray) positional variance, and has to work within 5° of angular variance (see Figs. 1 and 2).

High-intensity strobe lights supplied by DVT mounted below the vision window are synchronized to illuminate the target solar cell placed above the window at the same time camera exposures are made. The camera is also mounted below the vision window. Red LED modules from CCS America Inc. (Waltham, MA) mounted above the vision window provide contrast to the solar cell being inspected. A circular hood is used to reflect light downward—a technique Boston Engineering developed to improve image qualit while keepong the area clear for the robot arm.

The vision inspection system sorts nonconforming parts and places them into a dispensing tray for further manual inspection. Although essentially a go/no go task, sorting is based on a reference model in which at least 25 sensors are programmed on certain cell features. The word "sensor" refers to specific criteria programmed for target features using the software. When programming the DVT camera, these sensors are configured to look for certain desired features in the solar cell to within a defined tolerance.

The camera grabs and analyzes all the sensor information in one image and communicates the resultant commands to the robot (see magazine cover). For example, the commands may be analogous to a "Go ahead and place cell" or "Feature #24 out of tolerance, go to bin."

The positioning algorithm is implemented in a script file that outputs x-y and q command positions for the robot through a serial connection. The cell can be positioned based on two distinct cell features, and the operator has the ability to choose which positioning method to use via the HMI.

The FrameWork software can automatically calculate the amount of distortion in a given lens. The basic calculation devises a pattern of targets that are in known positions (this can be done in a word processor using tables of fixed cell heights and widths), uses soft sensors to find these targets, and applies the coordinate system soft sensor to calculate the amount of distortion.

Smithers adds, "We are trying to get to that ±0.01-in. spec, but have some minor distortion that is creating a ±0.020-in. maximum positional variance. The process is working, but we want to get that last 0.01 in. back, and that should be possible based on the sensor's specifications and with the support of DVT and Gibson Engineering."

Other advantages Boston Engineering found in the SmartImage Sensor are its ability to troubleshoot easily and to add more information through the Ethernet port. These changes are tested offsite through the SmartImage sensor emulator. Boston Engineering used the emulator to program the SmartImage Sensor, and then the inspection configurations were transmitted back to the manufacturing line via the Internet.

MICHAEL WILLIAMS is senior technical writer at DVT Corp., Norcross, GA 30093.

Allen-Bradley Inc.
Rockwell Automation Control Systems
Milwaukee, WI 53201

Boston Engineering Corp.
Waltham, MA 02451

CCS America Inc.
Waltham, MA 02453

Dumont Associates Inc.
Nashua, NH 03063

DVT Corp.
Norcross, GA 30093

Gibson Engineering Co. Inc.
Norwood, MA 02062

Intelligent Actuator Inc.
Torrance, CA 90505

Mitsubishi Electric Automation
Vernon Hills, IL 60061

More in Boards & Software