The present paper is a derived from an ongoing project oft he Chair of Measurement and Information Technology at the Helmut-Schmidt-University - University of the Federal Armed Forces Hamburg and the Department of Electrical Engineering oft bbw Hochschule - University of Applied Sciences Berlin. While the paper starts with a project level statement, it turns out to be more of a position paper with focus on two main problems, which shall be discussed in more detail; the results - in their simplicity - might have fitted into a ’Key Lessons Learned’ session as well.
Some oft he most delicate parts of turbines, especially in aircrafts, are the so called air blades. Their frequent inspection is an essential part of maintenance and a safety issue of highest importance. An inspection without disassembly is typically conducted using endoscopes respectively borescopes. Traditionally, the inspection is proceeded manually, by an experienced operator visually inspecting and deciding about the state of every single blade. This includes the choice of the field of view employing the limited degrees of freedom. The turbine is rotated manually or by using an high torque electrical step motor, the blade is brought into an acceptable angular position and inspected. There is an obvious need for certain aspects of automation in the procedure. It is desirable to store images of every single blade and track their history. Therefore such images have to be comparable. This requires a standardized process of image gathering, which first of all means constant and repeatable recording conditions. Some basic problems arise:
• In case the engine is inspected in the field, increased temperature has to be taken into account.
• Geometrical distortion of the image field versus their tilted main surface.
• The illumination conditions have to be such that images over the history can be compared, although the surface oft he blade might be subject to drastic changes in texture, roughness and color.
• The blades must be identified individually.
• The position oft he blade in the image has to be well defined.
Partial problem identification
Inspection hard-ware vs. temperature
Increased temperature inside the engine might be a problem for electronic cameras, which excludes proximate camera techniques and requires the standard passive distal borescope. The traditional use of passive distal borescopes is not just necessary for their robustness against high temperature above 100°C, but also for the fact that approved equipment which the operator is used to might be upgraded without excluding the operator from the traditional manual method.
Field distortion vs. correction
Figure 1 left shows the field distortion using an equidistant test pattern. Figure 1 right illustrates a typical viewing situation in one stage of a turbine. While the images is taken without the outer shell, in realistic situations there is limited degree of freedom for the borescope to be positioned.
Image field distortion as such is easily corrected; in conjunction with the tilted object surface it requires exact positioning, especially for objects to be stitched together from a multitude of single images. A correction of the tilted surface according to the Scheimpflug rule1–4 seems not useful for the fact that the tilting angle is not constant versus the rotation angle. A camera with adaptable sensor angle would increase the complexity to a level not suitable for field use.
Radiometrical calibration vs. illumination
A radiometrical calibration of the imaging chain is necessary due to the fact that images for a given object have to be comparable not only geometrically, but also concerning a possible drastic change of the surface and its reflectance. This requires consideration of two main issues,
While calibration and control of the source are of highest importance, standard solutions apply, which shall not further be considered in the paper. The question of dynamic range conversation leads the problem of the illumination geometry and the type of surfaces under inspection.
Dynamics vs. specular reflectance
The illumination problem is mainly a result of the fact that a borescope illuminates in the principal direction of view, which for shiny metallic surfaces leads to a hot spot in the direct reflex, while an oxidized or otherwise roughened surface will produce diffuse reflection. The situation is illustrated in Figure 2. In case a shiny surface is inspected, the image drastically looses dynamic range and tends to saturate, leading to a main issue of highest priority.
Positioning vs. drive
The problems of identification is strongly related to the positioning problem. If the inspection is proceeded on the turbine in its technical environment, say in the aircraft, there are a multitude of devices to be driven mechanically by the turbine. This leads to a rotation behaviour with the character of coupled rotational spring-mass-systems or pendulums with instable positions as known from synchronous motors. Especially at low rotational speed, there is often no quasistatic rotation possible. The turbine might be rotated in one main direction, but cannot be kept static in every angular position, neither manually or by the highest torque from a motor, but will find its next stable position.
Using high speed imaging at 500fps, Figure 2 shows example images, this unstable turning could be proved, even for speed values of several blades per second: there is a monotone increasing turning angle, but its derivative is not constant, below a certain minimum speed actually oscillating behaviour might be expected. On the other hand, the mechanical path usable for turning the machine does not give any hard reference to the absolute angular position of the blades which could be used to trigger the image acquisition. Both, absolute and relative positions, can only be gathered from the optical path. While the absolute position shall not further be considered in the paper, the relative position is critical and main issue.
In Figure 3 the consequences of the specular reflex for the image statistics is illustrated using a typical turbine blade out of its environment, on the inspection table. A first idea is to avoid and/or attenuate a multiple reflection between borescope aperture and sample. Figure 4 illustrates the various situations under consideration. The critical viewing situation with illumination and observation being perpendicular to the surface cannot be disclosed completely for every type of engine and blade, therefore the illumination angle is not a preferred parameter for optimization. The multi-reflection problem turned out to be of insignificant importance compared to the primary specular reflex. Increasing dynamic range of the camera to be able to equalize the spot numerically leads to dramatic waste of resources: Figure 5 quantifies the dynamic range necessary to gather the image of the source, in this case approximately 6 decades. While the LASER diode might be a hyperbolic model for a broad fiber bundle, even a broad spot being a decade above the target surface image will lead to artifacts when corrected by techniques like local equalization. An ideal solution for the inspection of shiny metallic and oxidized surfaces was found with cross-polarizing technique - in contrary to simple use of polarizing filters to exclude dielectric reflexes. This does not just lead to an acceptable attenuation of the direct reflex, but also to extremely improved colors. The shiny metallic surface becomes radiometrically and colorimetrically comparable to the oxidize one. Interestingly, the traditional visual inspection also suffers from the specular reflex blinding the observer. Although cross-polarization is nothing new, it seemed not to be introduced or accepted in the intended application. While in an idealized case in cross-polarization mode - non-polarized white light source - 50% loss by first filter, 100% diffuse non-polarized reflection - 50% loss by second filter - only 25% of the incident light might be used. In real situations less than 10% illumination efficiency should be expected, depending on the type of pollution of the surface. This results in strongly increased need for illumination power. If the scene is supplied by white light via fiber bundles, which is mostly the case in borescopes, the polarization filter is to be applied at the illuminating aperture. To deliver the light power needed without damaging the polarizing filter in the illumination path, short time illumination by flash lights or pulsed high power LEDs should be used, which has to be synchronized to the blade position.
Also, the polarization filter for the imaging path is to be applied at the aperture. A solution was found with an attachable device applied to the borescope according to Figure 7 left, which allows upgrade of existing equipment, but also quick change to manual mode - which, by the way, is also improved by the cross-polarization technique - and maintenance of bleached or otherwise damaged filters.
POSITIONING AND IMAGE ACQUISITION
From a certain average speed on the rotational angle increases monotone, the images have to be gathered from the rotating device. Then it is necessary that at least one image of every blade with constant position of the blades image in the field exists. Of course this can be realized with a video sequence. A short quantification shall set the frame for the timing. We assume a blade width b = 20mm and a minimum defect size Δe = 0.1mm to be detected with a spatial uncertainty Δs = 1mm or 5% of the blade width. A realistic minimum inspection rate of 1 blade per second - as a starting value which easily increases due to the conditions described above - then leads to a maximum exposure time Δt = 5ms, when a blur in size of the defect is accepted, equivalent to 200fps frame rate. The position requirement leads to a frame rate of 20fps, so in principle a 25fps cam employing exposure times < 5ms should deliver a sufficient video stream. The problem is in the unpredictable blade speed, leading to miss or hit images in a asynchronous sampling process. Ten times over-sampling, equivalent to 200fps, would be a consequence, requiring an immense data volume to be recorded and processed, which leads to increased costs at higher geometrical resolutions. Therefore we suggest synchronizing the image acquisition to the periodic scene which the turning blade wheel delivers. Then only one image is to be gathered of every bade at a relatively low average frame rate at short shutter times, so standard camera equipment delivers high resolution images with low blur. For illumination short time or pulsed sources are used anyway, due to decreased filter degradation.
In the prototype, the synchronization signal is derived from an optical position sensor at the object sided end of the borescope according to Figure 8, working in NIR, so there is no interference with the images acquisition due to the IR cutoff of the camera. In principal, other sensors, like ultrasonic, inductive or RF based devices might be used. To estimate the current speed, a second angularly shifted sensor might be used according to Figure 9. The blades deliver individual sensor signatures, which have to be taken into account in relation to the thresholds used. The process of gathering a whole series of one stage of the engine then - simplified - involves the steps:
• calibration run with signature to blade assignment,
• calculation of delay times for individual image acquisition,
• acquisition of final series.
In a next step, we intend to split an IR path from the VIS image path and use lower resolution second cam at higher frame rates for gathering the synchronization signal.
The synchronization involves illumination as well, due to the fact that at constant illumination - beside the energy consumption due to increased power for cross-polarization - the filter life time is drastically limited. So there is an interesting synergy between the solutions for the two main issues identified.
CONCLUSION AND OUTLOOK
In a short paper a concept and first results for an endoscopic inspection system was introduced, which might be used for maintenance of aircraft engine in garage and field, as well as for turbines in other assemblies, like power stations, compressors and the like. The basic problems identified lead to synergetic partial solutions, which deliver a robust system, designed for extension of existing hard-ware and keeping traditional inspection methods possible. Further investigations will involve the images processing procedures for defect detection and classification and refinement of the synchronization process using on-the-fly image processing.