Trends in x-ray casting defect recognition: advancements in assisted defect recognition can improve X-ray inspection in metalcasting facilities for consistent results and shorter inspection times.
The first production-approved, inline ADR systems for automotive castings were installed over 20 years ago, and this technology continues to be used. In many cases, this automation means there are no operators interpreting the images. These systems typically make an "accept" or "reject" decision based on the parameters supplied by a quality engineer and are designed to enable repeatable, reliable and documented results independent of human inspection error.
The acceptance and implementation of digital radiography (DR) in virtually every market has opened the door for many of the software tools, image processing and automation solutions made possible with digital imaging technology. Casting defects previously difficult to see are now easier to identify, and with the appropriate software tools, systems can help inspectors find, characterize and disposition anomalies and automatically accept or reject the casting based on the system settings (Fig. 1).
The terms automatic and assisted in ADR have different meanings. Automatic means no operator is required-inspection is fully automatic with fully automatic decision-making without operator intervention. Assisted means the system processes the images and indicates potential defects to the operator, who then uses this information to make the final decision.
[FIGURE 1 OMITTED]
The system is taught or can infer what a defect is, and the software analyzes the X-ray image and makes a decision or recommendation using characterizations, measurements, or other factors.
The goal is to provide more consistent and reliable evaluations by removing human variability and reduce casting inspection time by replacing manual inspections with automatic or assisted evaluations (i.e. therefore reduces part cost). This also helps address the scarcity of inspectors by reducing workload.
[FIGURE 2 OMITTED]
Metalcasting facilities can implement these automation and software tools in simple to complex systems with lots of variation in between. Typically, the transition to using digital imaging and software tools starts with manual inspection. In this mode, the system stops at every position and the operator evaluates the image and makes a decision. In some cases, the images are automatically acquired and stored for future evaluation.
Once the digital imaging and automation has been accepted at the plant and everyone is comfortable with the results, a common next step is to use some kind of semi-automatic or assisted inspection. In this mode, the system stops at every position, the software marks anomalies based on parameter settings for each view, and an operator makes a decision using this additional information. During this stage, the organization typically uses the data to begin correlating results with the manual inspection mode in order to gain confidence that the software tools are successful in helping the operator make better and faster decisions.
Once the software tools have proven to be effective (to whatever level of confidence is required), the next step is to start operating in a supervised automatic inspection mode. In this mode, the system typically involves the human inspector only if the casting has perceived defects. When the software detects something suspicious, the areas are marked and the operator reviews images and either confirms the software decision or reclassifies and overrides it. This step is critical to proving if the application is suitable for fully automatic defect recognition operation.
[FIGURE 3 OMITTED]
In many cases by this stage of implementation, the automated inspection system has met the goals of more reliable evaluations, reduced inspection times and reduced workload for the inspectors. At this point it becomes evident whether the application is suitable for fully automatic ADR, where the system is making both accept and reject decisions automatically without any operator intervention. This mode of operation balances the risk of the system missing an indication and the cost incurred by falsely rejected castings. To address this balance, some users use a remotely stationed operator who eliminates any false rejects by reviewing all suspect images at a networked computer. This is especially effective if the operation is running several systems simultaneously.
This transition can be tough, but the return on investment is normally worth it. Organizations that identify a champion who manages and teaches the ADR system tend to be much more successful with the technology than those that view it as a simple plug-and-play solution.
[FIGURE 4 OMITTED]
Different markets and applications have a wide range of requirements, but these current ADR systems can typically detect casting flaw sizes down to 0.1 sq.mm with a minimal depth of approximately 3% of the material thickness at an average 2-5% false reject rate. Of course, these values are strongly dependent on the application. Size/geometry, complexity, surface roughness, and whether the casting is finished or unfinished all have an effect on the inspectability of a part.
The detectable flaw size depends on the focal spot size, detector resolution and magnification-not the software. Critical for the results (especially the false reject rate) is the material and the geometry of the inspection parts. It is important to note, without a part-specific application it is difficult to estimate the quality of the ADR.
However, in every application the overall goal is the same as manual operator interpretation: To have zero false accepts. In order to assure this with an automatic system, falsely rejecting some parts is all part of risk management. Once the system is proven to be equivalent to the approved methods for never accepting a bad part, the system can be tuned to keep the false rejects at a minimum or even zero in some cases.
No matter what mode the ADR software is operating in, the results will depend on:
* Precise part fixing and/or manipulation.
* Part complexity and design, including surface roughness and production consistency.
* Optimal image quality.
Specific and well-designed software algorithms for image processing and defect recognition, as required by each application.
Maintaining software to account for production changes such as tooling wear, material variances and casting differences.
As an example, a common specification for defects in cast aluminum automotive parts is 0.5 mm. The image acquisition and evaluation process is optimized for a high throughput and low false reject rates. This is a calculated compromise between inspection quality and throughput.
[FIGURE 5 OMITTED]
A typical expectation for ADR (low false reject rate and high throughput) for automotive aluminum castings is 4% or lower. This application highly depends on geometry, part configuration and especially surface roughness. To help with the inspection, the software is capable of recognizing regular structures. A regular structure is defined as a reproducible geometric structure within a certain tolerance. These structures are taught to be accepted by the ADR software. After optimizing the imaging, and teaching the regions of interest with the appropriate filters, a false reject rate of even 2% is achievable in many applications.
At the most basic level, image processing for ADR involves several steps. All are equally important for the overall process.
Acquiring the image is the first and most important step and is crucial for obtaining the best results. A technique previously used for film or even digital radiography might not be optimal for an ADR implementation. Selecting the correct imaging chain with the right calibration and spatial resolution is critical. Choosing the correct magnification and integration time will define the detectability of the smallest defects. All of the details regarding focal spot size, energy, detector type, pixel pitch, scintillator material, geometry, and integration time must be optimized for these types of critical inspections.
Filtering the images is equally important because it allows the software to take advantage of the 16-bit images and find the smallest details. The filters are specific to the inspection task(s) and are designed to detect specific features (Fig. 2). Many defects that are difficult or even unable to be seen by the human eye can be automatically identified.
After filtering, the images are further processed for visualization and classification purposes (Fig. 3).
Filtering and processing also includes eliminating edge artifacts and pseudo-defects that cause false rejects. To avoid this, unique filters are applied depending on the part structure (Fig. 4).
In addition to detecting defects or anomalies, the defect recognition software must also classify the flaws. Flaws fall under many different categories and types, and each application has different requirements for accepting or rejecting the parts, for example (Fig. 5):
* Single flaw.
* Flaw distance.
* Flaw density.
* String flaw.
Once the entire image processing and classification has been completed, an "accept" or "reject" decision/recommendation is provided by the system. Limiting the defect recognition to an "assisted" mode will aid in identifying false rejects and confirming real defects, but the final goal (even in the most critical applications) still can be fully automatic processing. It may take some time and effort to create the correct algorithms, correlate the required data, prove the statistical approach required and then process the data needed to make this step, but the technology exists to make it possible.
[FIGURE 6 OMITTED]
In an existing investment cast part for an aerospace application, a test piece was designed and constructed with a range of small drilled holes. The inspection criterion is that 0.1-mm defects must be detected with a decision threshold of 0.2mm.
The "magenta" indications in Figure 6 identify the holes detected at 0.2 mm and above. These defects could either be rejected automatically or flagged to be reviewed in assisted mode.
The "green" indications identify the holes detected smaller than 0.2 mm, down to the required 1 mm size.
JASON ROBBINS, YXLON, HUDSON, OHIO
|Printer friendly Cite/link Email Feedback|
|Date:||May 1, 2016|
|Previous Article:||The next steps in metalcasting development through emerging technology, ideas: a sampling of presentations from CastExpo in Minneapolis focused on...|
|Next Article:||Improve results.|