Introduction Accurate and representative bubble sizes and distributions are used to characterize biochemical processes containing gas-in-liquid dispersions, specifically processes for industrially important fermentation products. Quantification of bubble sizes and distributions during fermentation is important to establish mass transfer characteristics (based on gas–liquid interfacial area) when oxygen transport to cells across gas–liquid interfaces becomes a limiting factor. In these situations, there is a direct influence of bioreactor parameters that affect bubble size, such as agitation, on culture yields. Thus, measurements of bubble sizes and distributions are useful for biochemical process optimizations. 1 2 Novel in situ probe measurement system EnviroCam™ probe 1 Fig. 1 Environcam™ shroud diagram  Various prototype shrouds and camera modules were developed to evaluate different hardware options. A high-strength sapphire window was integrated into the 316 l stainless steel shroud via a gold brazed joint rather than epoxy to provide improved robustness. The sapphire window was constructed at a 75° angle to offset the 15° angle of the fermenter’s Ingold port. This angle resulted in the window being parallel to the vertical tank wall upon insertion. Owing to the low surface tension of sapphire, some small bubbles adhered to the window surface, particularly at the low agitation speeds used for animal cell cultivation. Using the 75° vertical orientation, as well as raising agitation speeds, reduced but did not eliminate this accumulation. A ceramic disc backscreen provided a measured path length. Initially, light from a 150 Watt halogen lamp was reflected internally into the shroud backscreen via a high performance liquid light guide, but this arrangement did not result in uniform illumination. Next LEDs (red–orange 640–720 nm wavelength) were incorporated directly into the shroud backscreen for more uniform illumination, which produced bubble images with improved border definition. To minimize the impact of distortion in the depth and width of the measurement volume, a diffuser was installed in the backscreen. A calibrated reticle, consisting of two intrusion lines (180° apart with a gap of 7.5 mm), emanating from the circumference and heading towards its center, became the standard for in situ calibration. This reticle was located on the sapphire window itself rather than the backscreen to avoid interferences in opaque media. 1 EnviroCam™ imaging system 2 2 Fig. 2 Environcam™ system diagram  The number of pixels [horizontal (H) × vertical (V)] was altered using user-defined region of interest (ROI) controls. Higher pixel numbers increased resolution but reduced the number of frames per second (fps). For the selected CCD camera, the hardware was set at 20 fps using 1,024 × 1,024 pixels with a minimum pixel size of 6.45 μm and a pixel depth of 8 bits (without additional magnification via screw-on lenses). This fps rate of 20 resulted in a time scale of about 25 s for the initial image scan of 500 frames, which was small relative to expected changes in bubble size characteristics during the time course of a typical fermentation. Bubble residence time through the measurement field (1/2 in. = 12.5 mm path length) was quantitatively estimated to be about 1.3–3.3 ms near the impeller blade tip (impeller tip speed of 3.8–9.4 m/s) and likely was up to an order of magnitude slower away from the impeller. Consequently, it was not necessary to increase the frame speed further above 20 fps, which corresponded to a characteristic measurement time of 50 ms/frame. The bubble residence time in the 1/2 in. notch measurement field was qualitatively determined to be 40–70 ms by comparing common features of subsequent frames. Thus, it was necessary to skip at least four to five frames to ensure bubbles were not counted more than once. 2 3 3 2 3 3 6 7 2 2 The operating temperature range was limited to 0–50°C for the camera module; consequently, it was not attached to the shroud during vessel sterilization. The shroud LED-operating temperature ranged up to 80–90°C, but its non-operating temperature ranged up to 120°C. All other shroud components were steam-sterilizable, including the glass diffuser. Thus, the shroud could be sterilized with the vessel if the LEDs remained unpowered. A LED-power supply kill switch based on a bimetallic temperature sensor was installed with a trip value of 90°C and reset value of 60°C. When it was powered, typically intermittently for 10 s per frame measurement cycle or continuously for video stream, the LED was required to be submerged in liquid as a heat sink for adequate cooling. LED lifetime was 100,000 h assuming the non-operating temperature remained less than 120°C; raising it a few degrees above this level for effective sterilization might sacrifice some lifetime, however. Shrouds were heat-tested using a 15–30 s temperature ramp from ambient to 130°C, held for 1 h, then returned to ambient temperature. No significant degradation, as measured by pixel light output, was observed after 50–60 temperature cycles. In addition, an actual sterilization was conducted successfully with the shroud in a pilot scale fermenter (180 l volume, 122°C, 40 min hold time). Based on this performance, shrouds were expected to withstand about 100 sterilizations of 45–60 min hold times for about a 3-year life span, assuming a 2-week batch length. The 1951 USAF resolution target was used to evaluate the pure video resolution of the computer monitor, which was influenced by the quality of the video graphics card. Resolution was measured at 32 line pairs/mm, the reciprocal of which resulted in a resolution of 31 μm/pixel, significantly higher than the camera resolution of 7.5 μm/pixel. Thus, the accuracy of the image display did not diminish the accuracy of the photographs obtained. 3 2 8 9 Edge enhancement techniques, based on a contrast threshold, were applied to convert grey images to binary black and white images so that the outside perimeters (or diameters) of bubbles were readily identifiable via object recognition. 3 3 Green circles represented the first level of filtering. Rules in this first tier were: single bubbles of sizes within the target measurement range, circularity cut-off based on a user-selected tolerance above 1.0 value, and discard of bubbles touching the border. The green circle was the best fit “circle” so there were slight inaccuracies around some of the circumference if the bubbles were not uniformly round. Larger or smaller circular bubbles outside the target measurement range were excluded. Yellow outlines indicated the second level of filtering typically using similar rules as the first level but with more relaxed, user-defined cut-offs. Additional rules also were added in this second tier. Up to three, more or less circular bubbles, that were touching but still individually discernable, were included in the count. Specifically, the ratio cut-off of cluster (i.e., more than one adjacent bubble overlapping) area to the calculated equivalent area for completely separated bubbles of the cluster was based on a user-selected tolerance below 1.0. Bubbles in this category were further differentiable using additional capabilities included in the LabVIEW software. Red outlines comprised a third level that indicated bubbles located by the system that did not pass filtering levels 1 or 2. White outlines comprised a fourth level of classification. This fourth tier contained non-circular blobs and large irregularly shaped or greatly overlapping bubbles. The percentage of viewing area occupied by blobs was estimated and measured frames were omitted based on a user-selected cut-off (e.g., if greater than a target percentage of 75% of the measurement area was comprised of blobs, the frame was discounted). 2 The front panel image containing the configuration and user-adjustable controls was archived so that an identical set up could be reproduced in the future, if desired. Tiff files (still frames) and/or “avi” (video stream) files were archived, but avi files were reconstructable from Tiff files to save storage space. A video playback rate of 5 fps appeared appropriate. The actual measurement frames utilized for analysis were saved separately for ease of review. Minimum storage requirements for these unanalyzed frames were estimated at about 50 mB for each 500 frame measurement cycle (versus 4.5 GB for analyzed frames) with the entire analysis reconstructable from saved data. Thus, the ability to retrieve images and recheck/reanalyze results readily existed. 4 10 11 Fig. 3 Example main screen display  Fig. 4 a b c d e f  1 2 Table 1 Envirocam™ characteristics Camera and mode of attachment Magnification/calibration Illumination/shutter or frame speed Image measurement and analysis method CCD monochrome camera (Prosilica) probe inserted into an in situ Ingold-fitting shroud; 32 line pairs/mm 30×/internal reticle LEDs in back screen (back lit)/20 fps; 1/100,000 s (1/25,000 = 40 ms selected) National Instruments’ (Austin, TX) LabVIEW Graphical Development Environment as basis Measurement time per condition Number of objects per measurement Measurement error/size range Number of images (pictures/frames) per measurement 5–25 s data acquisition; <2 min data analysis >500 (typically up to 10,000) <10% for monodisperse beads/60–2,000 μm 50 (20–200 bubbles/picture) Bead calibration 3 3 3 12 13 7 2 14 15 3 4 7 Table 2 Distributions used to fit calibration bead measurement data (from Sigma plot software) Distribution Probability density function Modified Gaussian 4-parameter (G) Y Y o a  X X o b 2 Sigmoidal 4-parameter (G) Y Y o a X X o b Table 3 Comparison of manufacturer size (Whitehouse Scientific, Chester, UK) with measured size range for monodisperse beads Manufacturer’s size data (μm) Manufacturer’s data for 90% of beads within given range (μm) EnviroCam™ measured size (μm) EnviroCam™ data for 90% of beads within given range (μm) 22.81 ± 0.78 21.46–24.23 Not able to be measured  38.38 ± 0.54 36.5–39.6 Not able to be measured  59.63 ± 1.0 57.1–62.2 66.3 ± 0.05 (G) 58.8 ± 0.03 (S) 53.4–64.3 (S) 83.43 ± 0.87 79.7–87.5 82.5 ± 0.11 (G) 74.9 ± 0.09 (S) 64.2–85.2 (S) 98.10 ± 2.8 94.4–102.8 98.28 ± 0.82 (G) 90.64 ± 0.27 (S) 76.0–105.4 (S) 155.8 ± 1.5 151.4–163.1 163.9 ± 0.21 (G) 156.04 ± 0.2 (S) 143.7–167.4 (S) 200.9 ± 1.9 196–206 200.69 ± 0.49 (G) 194.15 ± 0.07 (S) 187.6–200.4 (S) 258.6 ± 5.9 251.4–265.6 259.77 ± 0.16 (G) 253.3 ± 0.07 (S) 246–261 (S) 297.9 ± 3.9 289.7–309.3 303.1 ± 0.05 (G) 295 ± 0.07 (S) 275.3–316.3 (S) 361.6 ± 9.9 344–376 366.5 ± 0.3 (G) 359.3 ± 0.86 (S) 337.4–380.8 (S) 405.9 ± 8.7 396–419 411.7 ± 0.07 (G) 403.9 ± 0.2 (S) 385.9–420.2 (S) 589.0 ± 6 572–615 586.7 ± 0.4 (G) 578 ± 0.09 (S) 551.9–606.6 (S) r 2 Table 4 Comparison of manufacturer size (Whitehouse Scientific, Chester, UK) distribution with measured size distribution for polydisperse beads Bead distribution (μm)  Size at fixed percentiles (μm) 10 25 50 75 90 Standard EnviroCam™ Standard EnviroCam™ Standard EnviroCam™ Standard EnviroCam™ Standard Envirocam™ 500–2,000 796 ± 10 744.5, 746.8 936 ± 9 849.3, 852.0 1,098 ± 11 1,069.6, 1,074.0 1,335 ± 32 1,232.0, 1,250.0 1,618 ± 40 1,442.5, 1,508.0 150–650 244 ± 4 245.8 306 ± 5 304.6 362 ± 5 368.2 424 ± 4 423.8 527 ± 18 493.9 50–350 94 ± 3.5 104.7 119 ± 1.2 130.9 151 ± 2.5 166.5 190 ± 4.8 202.8 237 ± 6.0 252.8 Envirocam™ measurements fitted to sigmoidal 4-parameter equation using Signmaplot software. Duplicate runs executed for 500–2,000 μm distribution Gas bubble data analysis The smallest bubble measurable was 30 μm in diameter, based on an expected maximum system resolution of 7.5 μm/pixel. A minimum of 2 pixels were needed to quantify the radii, and radii were used for diameter calculations based on initial ease of programming. Since the radius accuracy was ±1 pixel, the relative standard error for a 30 μm bubble was 50%, dropping to 25% for a 60 μm bubble. If required for other applications, smaller objects down to 20 μm might be measured using (1) two pixels to determine diameter or (2) using a back-calculated diameter based on equivalent surface area since only 2 pixels were required to define surface area. Bubble diameter measurements, generated using the image analysis software, were compared with those generated manually for 225 objects from a single frame image taken using the CCD camera system and a 1/2 in. gap shroud. Results demonstrated that the greatest percentage error was observed with smaller bubbles in the diameter range of 38–57 μm. This error generally decreased with larger bubble diameters. Manually measured diameters were slightly longer than image analysis measurements greater than 99.7% of the time, most likely due to a small amount of shadowing around bubble edges. The measurement error was 0.1% for the calibration line itself. Coincidentally, for some actual bubble measurement conditions, the number of objects rose considerably for bubbles 38 μm in diameter. This rise may be caused by the greater measurement inaccuracy at this size; thus small changes in size were not detectable. 3 4 7 D α N −1.2 2 2 2 5 5 Fig. 5 a b c  2 Measurement reproducibility was evaluated using bubble measurement data from a 15,000 l fermenter by analyzing every tenth frame for 500 frames starting at the first frame until 2,000 bubbles were obtained, then re-analyzing every tenth frame starting from the second frame, then again, starting from the third through tenth frames. The relative standard deviations of the averages typically were under 7.5% for the 90% cut-off Sauter mean diameter and under 4% for the 90% cut-off arithmetic and geometric means. Thus, the sampling of frames used for analysis was representative of the total number of frames collected. a a a 16 5 17 Test systems Water/media/broth 3 1 18 3 19 20 f 2 21 Amycolatopsis fastidiosa Fermenters equipped with open pipe/jet spargers When performing measurements in agitated fermenters equipped with open pipe/jet spargers (1 in. opening at 180, 600, and 15,000 l scales; 0.7 in. opening at 1,500 l scale), it was apparent that under certain conditions the bubble distribution was bimodal. Some images consisted of smaller spherical bubbles along with very large irregularly shaped “blob” bubbles, presumably owing to gas entrainment from vortexing and possible impeller flooding at higher aeration rates relative to agitation rates. Quantification of the discarded blob area from each of these conditions assisted in identifying the onset of flooding conditions, and possibly was directly related to the gas hold up (even under conditions in which individual bubble diameters cannot be discerned). 6 22 6 23 Fig. 6 Qualitative bubble size as function of agitation and airflow rate for open pipe sparger (1 in.) at the 180 l scale (0.001% P2000 in DIW)  7 23 7 7 D α N −0.45 D α N −1.2 2 Fig. 7 a b c  6 8 N Re, 4 5 Fig. 8 Comparison of bubble sizes in DIW and 50 vol% glycerol, containing 0.001% P2000, as a function of agitation and airflow rate at the 180 l scale  9 6 23 24 Fig. 9 a b  Fermenters equipped with ring spargers Bubble measurements in fermenters with open pipe spargers exhibited several bubbles in each frame, often in swarms. In contrast, bubble measurements in fermenters with ring spargers represented the opposite extreme in which only a few bubbles were present in each frame. These latter tests were conducted using purified water in 75 and 750 l geometrically similar bioreactors, spanning a tenfold size range. Fermenters were equipped with A315 impellers and a ring sparger with holes drilled 1/32 in. in diameter on the top surface of the sparger ring. Only a few bubbles were observed regardless of notch orientation (left, right, down, top). It was believed that the fewer bubbles observed per frame were due partially to lower gas hold ups, typically 0.02 vvm, but also the relative spatial placement of the probe at the level of the sparger ring. As the airflow rate was increased to its higher range values, more similarly sized bubbles were observed. In contrast, higher agitation rates caused more surface air entrainment; thus greater numbers of large, irregularly shaped bubbles were present in the frames. Higher airflow rates resulted in bubbles similar in nature to those observed at lower airflow rates, all governed by the holes in the ring sparger. In some cases owing to the slower agitation rates, smaller-sized (geometric mean of 95 vs. 200 μm) bubbles collected on surface of the shroud’s sapphire window, and these bubbles needed to be distinguished from the freely moving bubbles. Increases in silicone antifoam from 0.001 to 0.2% did not increase the low number of bubbles observed. Use of an in situ bubble measurement system in fermentation The ability of the Envirocam™ to measure in opaque solutions was examined using several model systems: For the 180 l fermenter equipped with an open pipe sparger, the effect of Pharmamedia on bubble images was investigated. When 20 g/l Pharmamedia and 2 ml/l P2000 was added to DIW (100 μs shutter speed, 20 gain, 185 brightness and 8 aperture), the contrast of bubble edges decreased to an unacceptable level. The particles caused granularity on the screen and blurred bubble edges, making detection difficult using the 1/2 in. gap shroud. When 20 g/l Pharamedia and 0.5 ml/l P2000 was added to DIW using the 1/4 in. gap shroud to decrease path length (100 μs shutter speed, 0 brightness, 31 gain, 8 aperture), bubble contrast was improved, but the resulting opaque solution appeared still too high for reliable analysis. Amycolatopsis  10 6 Fig. 10 a Amycolatopsis fastidiosa  b  10 6 Summary and future considerations A novel in situ bubble size and distribution measurement device was developed. The bubble measurement instrument design strategy permitted one camera module to be attached to a shroud, and thus one sensor (the camera) was able to be moved to multiple locations without disturbing the fermentation process. The small size and flexibility of the camera attachment permitted it to be readily relocatable. An Ethernet connection for the camera can further reduce the extent of the field hardware, and thus permit one camera system to be even more transportable. 2 At this time, application of the EnviroCam™ bubble measurement system appears limited to clear solutions that do not contain large numbers of overlapping bubbles. Different liquids (e.g., water, cottonseed flour, microbial broth, animal cell broth) possess different UV spectrum and light scattering properties, as well as varying surfactant properties, which influence bubble size and hold up. These differences suggest that some adaptation of the measurement system is necessary when moving from system to system. Specifically, agitation and aeration rate combinations which produce too many bubbles for measurement in a model DIW system may produce acceptable amounts of bubbles in a fermentation broth owing to changes in surfactant levels, but the ability to distinguish these bubble edges is diminished. Further expansion of the versatility and range of this instrument is the subject of future efforts, but key approaches being considered involve the further examination of available LabView filtering techniques to process bubble images and the use of smaller notch sizes.