X-ray Calibration for Void Measurement: Comparable Numbers

An X-ray system does not measure a void directly. It produces a projection image and a program decides where the void begins and ends. X-ray calibration is what makes that decision repeatable, covering the tube settings, the detector response, the geometry of the view and the thresholds used to segment the image.

Two systems can both be described as calibrated and still disagree on the same sample by several percentage points, because their thresholds differ. A void specification that does not state the method and the settings is therefore hard to enforce, and calibration is better treated as a production control than as an annual service item.

What Calibration Covers

Calibration covers four things: the source, the detector, the geometry and the software. The source contributes the tube voltage and current that set contrast and penetration. The detector contributes the conversion of radiation into gray level. The geometry fixes the magnification, and the software holds the thresholds that turn those levels into a void area.

All four drift, and they drift at different rates. A tube ages slowly, a detector loses gain unevenly across its area, and a threshold is changed by an engineer who is solving a different problem. Verification that covers only one of the four leaves the other three unobserved. The order in which they are checked matters less than checking all four against a written value rather than against a remembered one.

Tube Voltage and Contrast

The tube voltage sets how deeply the beam penetrates and how much contrast the image carries. Too low a voltage and the board absorbs the beam, so thick copper and dense packages appear as a bright block with little internal structure. Too high a voltage and the image becomes flat, and the difference between solder and void shrinks.

For a given package the setting is chosen once and then fixed, because changing it changes the measured void area. If the voltage is adjusted per board to make the image look better, the void percentages from that day cannot be compared with those from any other day. The same question of what a measured number actually means is examined in the notes on the void measurement method.

Detector Response and Gray Level

The detector converts the beam into a value for each pixel, and its response has to be uniform. A panel that is slightly brighter at the center than at the corners will report the same void as a different size depending on where it sits in the field of view.

Flat field correction handles the predictable part of that non-uniformity, and the correction has to be refreshed when the detector is recalibrated. Gain drift, dead pixels and a scintillator that has aged lie outside the correction, and they appear as a rising difference between systems inspecting the same sample.

Magnification and Dimensional Accuracy

Magnification is a geometric ratio, and it sets the scale of every measurement taken from the image. If the magnification is wrong, the void area is wrong by the square of the error, which turns a small geometric error into a large measurement error.

Dimensional accuracy is verified with a sample that has known features at the working magnification rather than by trusting the nominal value stored in the program. A sphere or a calibration pattern with a stated diameter gives the scale factor directly.

Void Threshold and Segmentation

Segmentation converts a continuous image into a binary one: solder or void. The threshold is the gray level at which that decision flips, and a change of a few counts moves the boundary of every void in the frame. Because the boundary moves, a void that sits near the limit can pass on one setting and fail on another.

Thresholds are set by the method, by comparison with a known sample, or at a midpoint between two peaks. Whatever the rule, it belongs in the calibration record, because a system recalibrated to different thresholds is not the same system even when the hardware has not changed. The joint defects that void limits are meant to catch are surveyed in the notes on voids in BGA joints.

Reference Samples and Their Limits

Reference samples provide a fixed target. A sample with a known void, measured repeatedly, shows drift in the system. Samples with features at several sizes show whether that drift is uniform or size dependent. A sample measured every month but never re-certified is a sample whose true value has quietly become an assumption.

The limits matter as much as the benefit. A reference sample measures the whole chain, so it detects drift without identifying its source. Its value is that it is fast, comparable between shifts, and can be run without an expert present.

Frequency and Triggers

Periodic verification on a fixed interval catches slow drift. Trigger based verification catches the events that matter more: a tube replacement, a detector repair, a software update, a change of magnification, a move of the machine, or a sample that produced an unexpected result. Every one of those events changes part of the measurement chain, and each is far easier to record at the time than to reconstruct afterwards.

A tube replacement is the clearest trigger, because the new tube has a different spot size and the geometry of the view changes with it. Recording the date of every such event against the verification result makes the later uncertainty shorter to resolve. The interval between scheduled verifications is a compromise between the cost of the check and the size of the drift that can pass unnoticed.

Comparing Systems and Sites

Where several systems or several sites measure the same product, correlation is a separate exercise from calibration. Each system is verified against its reference, and then the systems are compared with one another on the same sample. The sample used for correlation should carry features in the size range the specification cares about, because agreement on large voids does not imply agreement on small ones.

That comparison is what sets the tolerance a customer or an internal specification can use. If two calibrated systems differ by two percentage points, a specification with a one point limit cannot be applied across both without a rule for which system is authoritative.

Records and Traceability

The record of a verification should hold the date, the settings, the reference sample and its certificate, the result, and the action taken if the result fell outside the limit. Void percentages reported to a customer without those fields are an opinion about an image.

Traceability also means that a threshold change can be found later. Where the setting lives in a program file, the file version and the date it was loaded serve the same purpose as a written record, provided the file itself is under control. Where the same imaging chain also supports other acceptance decisions, such as barrel crack inspection, one record can cover both.

FAQ

Does X-ray calibration guarantee that void measurement is accurate? It makes the result repeatable and comparable. Accuracy against a true void size still depends on thresholding and on the geometry of the view.

How often should the system be verified? On a fixed interval for drift, and after any event that changes the source, the detector or the software.

Can two systems report different void percentages? Yes. Different thresholds and different geometry produce different numbers from the same sample, which is why the method has to be stated.