Quality and manufacturing engineers are in a continuing battle to increase their competitiveness by improving quality and throughput while decreasing the cost to manufacture. The challenge to find the most effective way to inspect product in-process can be overwhelming, with choices of photoelectric sensors, vision sensors and machine vision systems. Each provides great benefits to the manufacturing process, but when is one too much, and the other not enough to do the job? Finding that answer ahead of time saves time, money and lots of late night phone calls from the plant floor.
Inspection at the end of the production line is a great way for manufacturers to catch flawed products, prevent bad shipments and avoid rejected parts, costly recalls and product returns. However, waiting until the final check for a remedy after a product has undergone numerous value-added stages of production often creates waste and time-consuming rework. In addition, defects not caught at the source generate little information about their cause that can be used to improve the process.
To increase their competitiveness, manufacturers have made inspecting product in-process a major priority. Photoelectric sensors, vision sensors and vision systems have clearly emerged as the tools of choice for manufacturers in pursuit of increased quality and profitability through inspection, product tracking and process control. But these tools also are one of the least understood and most erroneously specified areas of automated manufacturing.
A variety of new inspection products from machine vision and sensor suppliers is providing standard offerings with advanced features required for many specialized inspection applications. The key problem is a lack of practical information about implementing these tools in real-world applications that is necessary to help sort through the confusing matrix of available technologies.
Sensing TechnologiesVarious manufacturing and product goals demand different sensing technologies. Industrial production relies on the automatic sensing of parts being manufactured. Three distinct forms of noncontact sensing: photoelectric sensors, vision sensors and machine vision systems, each with characteristic benefits and tradeoffs, are in widespread use. With all of the available offerings, there is sure to be a sensor product for any given application. But sorting through the wide array of products and datasheets can be a daunting task for even the most experienced.
The first step in specifying the most cost-effective solution to achieve specific manufacturing and product goals is to understand the advantages and disadvantages of each technology as applied to the application at hand. The most common uses for photoelectric sensors, vision sensors and vision systems are detection, synchronization, inspection and more complex tasks such as defect classification or optical character verification (OCV). These uses are defined in more detail below.
• Detection: Senses an object that has reached or crossed a defined reference point.
• Synchronization: Tells where the object is at a certain point in time so that automation equipment, such as a reject actuator, can be synchronized with the position of the object.
• Inspection: Determines whether the object satisfies certain inspection criteria, such as presence or absence of visible features, so that an accept/reject or similar decision can be made.
• Machine Vision: Includes guidance or alignment, gaging or precision measurement, complex defect detection and classification, and output data for process control and traceability.
From part finding triggers to complex defect detection and classification, manufacturers must balance the cost of implementing photoelectric sensors, vision sensors and vision systems in light of these application requirements.
It is always best to first decide on the type of output required for the application. Photoelectric sensors or vision sensors are typically the least expensive and easiest solution if all that is required is simple pass/fail binary output indicating whether a product is present or absent, good or bad. Vision systems offer the best solution when guidance, gaging, complex defect detection or output data for process control and traceability are required.
Optical and vision sensors provide simple pass/fail output. A wide variety of photoelectric sensors are available for many common detection, synchronization and inspection applications. Though low cost and easy to set up, these sensors can cause frustration when applied to manufacturing problems beyond their traditional capabilities. That is why machine vision and sensor suppliers have introduced a new inspection class of vision sensor products that combines the power of machine vision with the ease of use, fast deployment and high-speed binary output of traditional sensors.
Photoelectric SensorsPhotoelectric sensors do the bulk of industrial part presence detection. Photoelectric sensors are available in a variety of styles, sensing modes and performance levels, designed to address a host of applications. The size, shape, color, opacity and reflective properties of the object to be detected, as well as response time, mounting and other application- specific requirements determine the best combination of sensor components to achieve optimal performance in a given application.
These requirements also will determine the best sensor mode, which could be transmitted beam, retroreflective, diffuse proximity or other modes. No matter which sensor mode and components are selected, most photoelectric sensors incorporate operator-selectable sensitivity to establish a light intensity above which (or below which) an output signal will be energized.
Part Presence - Simple ApplicationsPhotoelectric sensors perform best when looking for the presence of an object moving on some form of a conveyor system. The most reliable method of detecting objects is to use a thru beam photoelectric pair, having a sensor on one side of the conveyor emitting light and having another, the photodetector, on the opposite side receiving, or detecting the light. When the leading edge of the object breaks the light beam, the sensor detects the presence of the object. While this may be the most reliable approach, it does have some drawbacks: clear objects are difficult to detect, and this setup requires installing, wiring and aligning two sensors for each application.
These drawbacks have led people to solve these applications in alternate ways while attempting to reduce the total cost of solving the sensing need. One of these solutions uses a single, retroreflective photoelectric sensor, which both emits and detects the light. It uses a reflector on the opposite side of the conveyor to bounce the light back to the receiving portion of the photoelectric. This design serves the same purpose as a pair of thru beam photoelectrics and eliminates the need to wire the second sensor. However, it still requires a reflector to be mounted on the opposite side of the conveyor and may still have difficulty with clear or highly reflective objects.
To further reduce the costs of installation, applications are being addressed by diffuse (or proximity style) photoelectric sensors. This design has a single sensor that emits light; the light is reflected off the target and back to the sensor. The sensor detects the reflected light to sense the presence of the object. This eliminates the need to purchase and mount the reflector, but this approach has drawbacks as well. Typically the part will need to have a highly repeatable position to be detected consistently. The shape, color, opacity and distance from the sensor as well as the angle of the target impact the ability of the sensor to reliably detect these targets. Objects in the background also may become an issue if the sensor detects these objects instead of the desired target.
Within the photoelectric market, these basic modes of operation are put into a variety of housing styles to allow them to be mounted in different ways to address different applications and environments. To further this, photoelectric designs are offered in fiber optic configurations, removing the electronics from the application area, reducing the size of the sensor that needs to be mounted in the application environment, creating greater mounting flexibility, increased environmental resistances and the ability to detect even extremely small objects such as ICs or capacitors.
Overall, photoelectric sensors work very well for traditional part presence applications.
Synchronization ApplicationsWith photoelectric sensors, detection and synchronization are essentially the same operation. While these sensors have been designed to detect a specific target, synchronization requires logic, which typically occurs in a programmable logic controller (PLC) or similar device to delay the output signal and synchronize a downstream operation such as web slicing or part reject diverter. Sensor repeatability is generally on the order of several hundred microseconds, which is generally acceptable for most synchronization applications.
For certain applications, different versions of photoelectric sensors have been designed to address the unique requirements of these detection- sensing applications. For example, for synchronization of manufacturing equipment involving web systems, photoelectrics have been designed to detect the presence of specific targets such as luminescence marks, labels on reels, web edges and color marks. These inputs are provided to a PLC, which then synchronizes the manufacturing equipment based on the timing of this input.
While the sensors can be reliable, they do introduce requirements that result in a more complex system and contribute to an increase in the total system cost by requiring: 1) a unique target to be added (such as the color mark) and 2) the creation and management of program logic, which is significantly more expensive than the purchase price of the sensor itself.
Inspection ApplicationsA basic photoelectric sensor is a part detection device. They have the ability to look at one point and indicate whether a part is present or not. If the sensor is used as an inspection device, a single sensor is looking for a specific feature on a part. In order for the sensor to determine if the feature is present, the application typically requires that:
• The object is fixed in a highly repeatable position and orientation during the detection process.
• The object has only one or two features to check that are not too close together.
• The object or features of interest are detectable through normal photoelectric sensing techniques.
• Changes in size, shape, color, opacity and reflective properties of the target object are limited.
In addition to detecting the feature, an additional sensor is typically used to detect the object the feature is on. These two inputs are then looked at to: 1) determine that the part is present and 2) determine whether the feature on the part is present. This logic is generally added to the PLC programming to make the ultimate feature sensing decision. This level of complexity, cost of installation and ongoing maintenance issues are driving the market to search for better solutions.
Because photoelectric sensors are fast and inexpensive, there are many applications where photoelectric sensors are well suited for part detection and some synchronization. When the application involves telling machinery whether something is there or not, photoelectric sensors may be the most cost-effective choice if the following conditions apply:
• The object is fixed in a highly repeatable position and orientation during the detection process.
• Changes in size, shape, color, opacity and reflective properties of the target object are limited.
• If the above conditions do not apply, consider moving up to an inspection-class vision sensor.
Vision SensorsVision sensors are more capable replacements for traditional presence sensing devices. Some detect and inspect parts on production lines or check for presence of specific features on products and packages. Others perform tasks such as synchronization, profile detection, object recognition or height measurement. No matter the task, vision sensors have proven to be very effective in overcoming many of the deficiencies of traditional photoelectric sensors when:
• Inspected points are close together, difficult or impossible for photoelectric sensors to detect.
• Precise part handling with mechanical fixtures is impractical.
• Mounting sensors close to the part is problematic.
• Multiple areas must be inspected on each part.
• Frequent changeovers are required for mixed-model processing.
The conditions above set the stage to illustrate some of the shortcomings of conventional sensors. For example, simple sensing of light intensity at a single point often is insufficient for many industrial inspection applications. Instead, it may be necessary to analyze a pattern of brightness reflected over an extended area to determine light-to-dark transitions to detect edges or patterns. Inspection-class sensors that incorporate a multi-pixel image array can more efficiently accomplish these types of tasks and detect features that photoelectric sensors cannot.
Another drawback is that objects presented to a photoelectric sensor or arrangement of sensors must be fixed in a known, predetermined position to ensure the appropriate points are inspected. However, the added cost of even rudimentary mechanical fixtures can frequently offset any savings from the lower initial cost. To accommodate positional uncertainty, most inspection-class sensors use a more expensive area array. The added cost is easily justifiable when it eliminates the need for mechanical limiters and precise part handling, or if manufacturing flexibility is desired.
Many suppliers also offer low-cost lens kits that allow vision sensors to provide easy working distance changes for additional mounting flexibility. This is especially useful in applications where it is not possible to mount the sensor in close proximity to the part and long-range inspection is required.
While inspecting many points on an object is possible with photoelectric sensors, it can be very complicated. Each sensor must be carefully mounted to avoid light-source crosstalk or other interference caused by space limitations. Each sensor must be wired to a PLC, and ladder logic must be programmed to combine the inputs from all of the sensors in order to make a decision on whether the item passes or fails the inspection.
To greatly simplify this type of application, some vision sensors do not limit the number of areas on a part or package that can be inspected, and include PLC-like capabilities to provide one pass or fail result without PLC or ladder logic. This typically requires that the vision sensor incorporate a processor and memory. Some more advanced vision sensors directly accept encoder input and include an automatic shift register to track and reject items on variable speed packaging lines. Finally, many vision sensors can store multiple setup files for various products that are selectable by discrete inputs, for fast line changeovers.
Photoelectric sensors, and arrangements of photoelectric sensors, by comparison, are very inflexible for manufacturers producing a mix of products, especially when each product has unique inspection requirements that make line changeover from one product to another a time-consuming manual process requiring sensors to be physically moved and readjusted. The cost of performing such a line changeover, and the risk of human error involved, often offset the low cost and simplicity of traditional photoelectric sensors.
It is important to note that for a vision sensor to detect and synchronize on par with traditional photoelectric sensors, it must provide internal part triggering and timing repeatability on the order of several hundred microseconds or less. Today’s most advanced vision sensors can deliver output timing repeatable to better than 100 microseconds. For precision detecting for cutting or applying, these fast vision sensors are far more precise than traditional photoelectric sensors.
Finally, one last advantage of vision sensors is that many allow manufacturers to specify certain types of images to be recorded during part inspection. This helps users to better understand their process during troubleshooting and identify why parts are passing or failing a particular inspection.
When more than simple pass/fail binary output indicating whether a product is present or absent, good or bad is required, vision systems offer the best solution for the following uses:
• Gaging: to measure a part or examine its critical dimensions.
• Inspection: to indicate if a part is good or bad based on its physical characteristics.
• Guidance: to accurately locate, align or place a part.
• Identification: to determine whether the right part is present by inspecting its physical characteristics or reading a marked code.
Vision SystemsVision systems provide process data that must be interpreted. Vision system hardware typically includes a camera that captures an image of the item to be inspected, lighting to enhance the contrast or features of interest and optics, which accurately represent the image to the camera by minimizing distortion and loss of resolution. This hardware works with a processor, or vision engine, to capture, digitize and display images for analysis, and to generate answers, such as whether a part is defective or to guide a process.
Vision software tools are the backbone of any vision system. A wide assortment of vision tools are available for performing many different types of inspection operations that enable vision systems to make decisions about a part’s quality, location, size and identity. This brief list describes some of the more common vision sensor tools:
• Part Location: Searches image to accurately and reliably find a part.
• Edge Detection: Locates edges in an image while ignoring variations in its background, and calculates the magnitudes and angles of edges.
• Caliper: Performs high-speed, sub-pixel measurement of the width of part features.
• Blob Analysis: Performs highly repeatable measurements of the area, size and centroid of part features.
• Calibration: Calibrates camera pixels to real-world engineering units.
• Nonlinear Calibration: Optimizes system accuracy, corrects for lens and perspective distortion.
• Multi-Pose Nonlinear Calibration: Optimizes system accuracy, corrects for lens and perspective distortion, and enables the use of a smaller, more manageable calibration plate in large field-of-view applications.
All these vision tools provide tremendous application versatility, which is one of the biggest advantages of using vision systems. However, running these complex algorithms blunts the suitability of vision systems for detection or synchronization due to slower and varying decision-making speeds on the order of tens to hundreds of milliseconds.
Like vision sensors, vision systems avoid the disadvantages associated with using an arrangement of photoelectric sensors. They can analyze patterns of brightness reflected from extended areas, easily handle many distinct features on the part, handle uncertain and variable part locations, and accommodate line changeovers through software.
However, unlike vision sensors that interpret the video automatically to provide simple binary pass/fail output, vision systems must be configured or programmed to interpret the process information generated and determine the result. Consequently, they are more difficult and costly to use and commonly require some level of specialized knowledge or training.
In summary, vision systems are most easily cost justifiable when used for control functions that require networking or a tremendous amount of process data for trend analysis, continuous process improvement, quality control, process control, machine control or robot control. V&S