Magnaflux Quasar Technology

Process Compensated Resonant Inspection

Magnaflux Quasar Process Compensated Resonant Inspection (PCRI) is a powerful, yet flexible NDT method that provides superior NDT at substantially lower cost. Quasar PCRI technology is built on 5 critical elements:

Limitations of Conventional NDT Methods

The purpose of production NDT is to reject defective parts - that is, parts that would fail prematurely in service. Conventional NDT methods, such as X-Ray and magnetic particle inspection (MPI), identify defective parts by scanning for indications of a defect. The assumption is that the presence and size of the indication correlates with the presence and severity of the internal defect. The problem with this assumption is that the physical phenomena that serve as the basis for the NDT method (e.g., the magnetic field in MPI) are not related to the mechanical performance of the part under test. This means that the apparent size of the indication is a weak predictor of part performance , so the parts rejected are not necessarily defective. This lack of correlation to part performance is common to all conventional NDT methods. A further limitation of many NDT methods is that they require the exercise of human judgment, which is inherently fallible. Independent studies have shown that humans, in the best circumstances, are at most 80% reliable in correctly detecting indications.

 Use of Mechanical Resonances

A part's mechanical resonances are determined by its material properties, which also determine its mechanical performance. So resonance testing offers the potential for performance-based NDT. A part's resonant frequencies are a direct function of its stiffness and mass. For example, the frequency of the first bending resonance, f, is:

                            f =  (1/2pi) * (k/m)1/2

Where: k is the stiffness of the part and m is its mass. A defect such as a crack, inclusion or porosity, reduces the stiffness of the part and therefore reduces its resonant frequencies. The change in frequency is proportional to the change in stiffness, so it is potentially a predictor of the part's performance.

Unfortunately, there is an obstacle to using resonances for NDT. Acceptable process variations also affect the resonant frequency to the extent that they mask the effect of even a fairly severe defect. This is illustrated in the figure below, which is a histogram comparing the resonant frequency for a sample of 200 good and bad (defective) cast aluminum master cylinder bodies. 

  ResDist.jpg

The bad parts in this sample set represent oxides, porosity and cracks. The good part frequencies vary over a range of about 1% because the dimensions vary over time as the process changes, e.g., cavity wear. The variation in the set of good parts is relatively small (sigma = 0.2% of the mean) because the process is in control. As a set, the bad parts have a lower average frequency and also have a broader distribution (sigma = 1.4%) because they represent an out-of-control process state. The average bad part frequency is lower because of the reduced stiffness, but a few bad parts have higher frequencies because the defect (a large shrink porosity, for example) reduced the part's mass. The vertical lines indicate a window that would accept all of the good parts. However, this window would only reject 35% of the bad parts. This is a typical result demonstrating that simple resonance measurements can only reject grossly defective parts.

Process Compensation

Magnaflux Quasar has developed a method that compensates for the acceptable process variation and effectively unmasks the defects. The Quasar method measures several resonances for each part and uses a proprietary pattern recognition algorithm to compensate for the acceptable process variations. This computation is made using the Mahalanobis Taguchi System (MTS) to describe "good" parts in terms of the measured distributions of a set of resonant frequencies. The MTS computation is then augmented with another more powerful (but less general) mathematical tool called a quadratic discriminator. This discriminator is used to differentiate between the general class of "good" parts, and one or more specific classes of defects. Together, the MTS and the discriminator usually provide 100% sorting of the defective parts.

The calculated pattern predicts the frequency of a target resonance for each part. The difference between the predicted and measured frequency for a part is its Predictor Error. Good parts have a small error; bad parts have a larger error. The next figure shows a calculated Acceptance Window using the relatively small error computed for the good parts. Parts with errors outside the Acceptance Window are rejected. As seen below, the Predictor Error readily discriminates between the good and bad parts despite the overlap in the raw frequency measurements.

PredErr.jpg
The compensation pattern is visually illustrated in the next figure, which shows the measured and predicted frequencies for each of the parts in the sample. Note that the Acceptance Window is shown here as an ellipse, but is actually a multi-dimensional ellipsoid, since several resonances are typically used (five in this case). 

MDEllip.jpg

Precise Frequency Measurements

The previous figure also illustrates another important element in the Magnaflux Quasar method - note the size of the data points. Each point can be thought of as including the statistical uncertainty associated with its measurement - that is the error bar. The error bar depicted here is 0.03%. The next figure shows the same plot with an error of 1% for each data point. As seen, the MTS window no longer provides effective sorting.

DPSize.jpg

This illustrates the importance of precise frequency measurements. Most resonance testers measure a part's resonant frequency by simply striking it and mathematically transforming the sound into a set of resonant frequencies. From a practical standpoint, this approach is limited to an accuracy of about +/- 1%, which (as seen in the figure) is not adequate for effective compensation. Quasar uses swept sine waves to measure the resonant frequencies. In this method, the part is driven by a sinusoidal vibration at a given frequency and its vibration is measured by a receiver. The driver then steps to the next frequency and the process is repeated as the driver sweeps across a designated frequency range. The swept sine method can achieve an accuracy of +/- 0.001%, which makes compensation feasible.

However, even the swept sine method cannot, by itself, deliver the required accuracy. Two additional factors are required.

The measure of frequency precision is called the margin, which is the standard deviation of the measured frequency over a statistically significant number of measurement repeats. The combination of the swept sine measurement, temperature compensation and precise fixturing provide margins that typically range from 0.003% to 0.3%.


Correlation to Mechanical Performance

The Quasar test generates a "Quasar score" for each part that is a predictor of its performance in the field. The figure below plots the Quasar score vs. the break force for a sample of powder metal exhaust flanges. As seen the correlation is almost perfect. So Quasar PCRI provides the ultimate NDT benefit - prediction of the part's field performance. So only truly defective parts are rejected, and these are rejected with near perfect reliability.

 

CorrMP.jpg

Rugged, Factory Ready Packaging

This discussion has presented the theoretical basis for the Magnaflux Quasar method. However, the theory is not valuable useless unless it actually works on the factory floor. Magnaflux Quasar has developed rugged hardware that can be integrated directly into the production line. It features NEMA enclosures, PLC controls, shock isolation and power line filtering. These systems are in place and testing production parts 24/7 in factories throughout North America. The figure below shows one such dual test station system configured to test powder metal rings.


FactSystem.jpg