3.1. Comparison of AE, Bubble Activity, and Atmospheric CorrosionFigure 5 shows the curve of the measured temperature T and RH during the atmospheric corrosion experiment, which lasted about 96 h. The temperature varied periodically with the day-night cycle; the minimum value during the night was 22 °C and the maximum value during the day was 25 °C. The initial RH was about 35%. The controlled variation started with a rapid increase in the RH, which approached about 77% RH at hour 2.8 (i.e., at the end of step 1). In steps 2 to 5, a controlled decrease to a low RH of about 33% at hour 23.5 was performed. In steps 6 to 10, the RH slowly increased again to a maximum RH of 85% at hour 76.7. Step 11 started with a rapid drop in the RH and then continued with a slower decrease to a final RH of 7% at hour 95.8. The times that the saturated salt solutions were exchanged, i.e., the changes between the steps, are indicated by vertical dashed lines and can also be seen in the abrupt drops or rises in the RH curve. Although the expected RH values, as given in Table 2, were not achieved for the saturated salt solutions used, the RH variations were suitable for simulating the dry and wet phases of the monitored specimen, as expected in the operation of aircraft components. Moreover, Figure 5 shows several images of the specimen’s surface with the deposited droplet (extracted from the recorded videos) and the corresponding time points. Figure 5a shows the droplet at 2.8 h after the start of the experiment, where a high RH of 77% was present. Bubbles of different sizes are clearly visible in the droplet. Due to the absence of other bubble-forming processes and the corrosion taking place (as can be seen later), this was determined to be H2 produced by corrosion processes. Figure 5b shows the surface of the specimen at hour 23.5, which was the end of the defined decrease of the RH to 33%. The steady decrease in the RH caused the water to slowly evaporate from the specimen’s surface, whereas the NaCl crystals of the droplet solution remained on the surface. Furthermore, the formation of pits can be clearly seen in the figure. Figure 5c shows the surface at hour 54.9 at an RH of 79%. Due to deliquescence, the moisture in the humid environment was absorbed by the NaCl crystals, which resulted in numerous smaller droplets of another corrosive NaCl solution. Figure 5d depicts the situation at hour 74.6 with an RH of about 83%, which was only briefly before the maximum RH of 85% was reached. Due to the increasing duration and RH, the droplets increased further in size and partly merged. Moreover, by comparing Figure 5b–d, the growth of the already existing pits and the formation of new pits can be clearly observed. Figure 5e shows a completely dried-out but corroded surface after 95.8 h at an RH of 7%. Significant pits and NaCl crystals can be identified. The minor pit growth observed between Figure 5d,e was also expected, but due to the short time (about 2 h) in which the electrolyte was present on the specimen in this period, this was not visually apparent. After the rapid decrease in the RH at hour 76.7, the pit growth stopped.Figure 6 illustrates that corrosion did not progress during the dry phases at a low RH, which can be seen by comparing the exemplary images of the corrosion site (extracted from the recorded videos) in these phases. Due to some issues with the video control, some periods exist where no videos were recorded, e.g., in the dry phase from hour 9.8 to hour 23.5. Therefore, Figure 6a shows the specimen’s surface only at hour 23.5, i.e., at the end of the defined dry phase, and Figure 6b shows the surface at hour 27.5 at an RH of 47%, where the specimen’s surface was also still dry. During this period, neither the pit growth, formation of new pits, nor changes in the deposits were identified. However, a closer look at the corresponding videos revealed that tiny bubble activity already started again at hour 27.5. Therefore, the dry phase at the end of the experiment was also analyzed. Figure 6c shows the dry specimen surface at hour 77.7 at an RH of 27% and Figure 6d shows the surface at the end (18.1 h later) of the experiment, i.e., the same frame shown in Figure 5e. No differences were found between Figure 6c,d. Moreover, no bubble activity was observed in the videos recorded during the corresponding time period. Thus, it is concluded that corrosion did not progress during the two dry phases of the experiment.To address the initial question of whether atmospheric corrosion can be measured by the AE and further, how it is related to the RH, the measured AE, video, and RH data were processed and analyzed to determine if correlations existed. Commonly, AE signals are analyzed by features such as the AE hit count, hit energy, peak amplitude, root mean square voltage, etc. [32]. However, for this fundamental investigation of whether atmospheric corrosion can be detected by the AE, the cumulative AE hit energy was found to be sufficient (see results below). An amplitude threshold of 28 dB with a reference voltage Vref = 1 μV [32] was used for the extraction of single hits from the continuous AE data stream. The threshold was chosen to be 3 dB above the noise level (proved to suppress small noise outliers), which was approximately constant at 25 dB for the entire AE data. Further, for all the detected hits, a 5 ms time signal was used for further analysis, i.e., a predefined hit duration was applied. The 5 ms duration was enough to fully record most AE events that triggered a hit. Although the experiment was conducted under laboratory conditions, electrical interference could not be completely avoided. However, such interference typically generated spike-like AE signals with high amplitudes but only very few (often just single) counts (number of positive threshold crossings) [32,41]. These interferences were filtered out by not considering AE hits with fewer than five counts.For the monitoring of the corrosion propagation, the videos were visually inspected (results are presented in Figure 5 and Figure 6). It was observed that the corrosion of aluminum and its constant production of H2 also formed bubbles when there was only a small amount of electrolyte present on the aluminum surface. These bubbles are reported to be one of the major sources of AE in immersion-like tests [26,27,28]; thus, the observed small H2 bubbles were believed to also be a major AE source in the present atmospheric corrosion experiment. To investigate this question, the recorded videos were processed using an algorithm implemented in Python 3.8 that automatically detects sudden changes such as moving, merging, or bursting bubbles. In the first step, background subtraction using the OpenCV package was conducted to highlight sudden changes in the videos. In the second step, event detection by evaluating the size of these changes using the imutils and OpenCV Python packages was performed. In cases when the detected events exceeded a certain threshold size, the events were considered to be relevant. For comparison with the AE data processing, a cumulative count of the detected events over time was generated. Figure 7 shows some exemplary results of a processed video recorded in step 1. The upper part of the images shows the originally captured frame and the lower part shows the corresponding processed frame. The original frame in Figure 7a shows several bubbles in the droplet, and few changes from the previous frame were detected, resulting in an almost black processed frame. Figure 7b shows the subsequent frame. The bubble on the upper-left edge of the droplet has burst, which was detected by the algorithm and is highlighted with a green rectangle. Additionally, two smaller events were detected in the lower-right corner of the frame.The authors decided that a threshold size of 60 pixels was a proper value to avoid the invalid detection of noise or other artificial objects caused by small vibrations or shocks to the setup. However, larger shocks to the setup caused several detections at the same time, which resulted in a heavily disturbed cumulative count (cf. Figure 7c). Such disturbances were removed from the cumulative count in a final processing step. The manual inspection of some processed videos revealed that there was predominantly H2 bubble activity in the videos and that this activity was well captured by the video processing algorithm. Furthermore, the manual inspection revealed that in most cases, a maximum of three to four valid objects were detected between two consecutive frames. Thus, if five or more objects were detected simultaneously, the corresponding counts were removed. In another cleaning step, a check was performed to see if an event had been detected in the same frame or a nearby position in the previous frame. This avoided the multiple counting of single events of longer durations, e.g., if bubbles slowly moved or rose.The normalized cumulative AE hit energy and normalized cumulative video event count are shown together with the curve of the RH in Figure 8a. The normalization of the two cumulative quantities was performed using their maximum values, which were 1.99 × 106 eu and 82,555 video events, respectively. As mentioned above, there were some periods where no videos were recorded, especially in the time span from hour 10.6 to hour 23.5, which is illustrated by the solid line without markers on the curve of the cumulative video event count in Figure 8a.Figure 8a clearly shows that the cumulative AE hit energy and cumulative video event count were related to the RH, or more precisely, to the resulting presence of electrolyte on the specimen’s surface. In the initial phase, when the RH rose sharply to high values so that the droplet of the NaCl solution remained on the specimen and did not evaporate, both cumulative quantities showed a significant increase. After about 3.7 h and an RH drop to 66%, which was after the second exchange of the saturated salt solutions, the curve of the cumulative AE hit energy significantly flattened. The corresponding videos showed that some water of the droplet had already evaporated; thus, less solution was present on the specimen. After about 7 h at an RH of 63%, the stagnation of this curve can be observed, whereas the curve of the cumulative video event count started to stagnate later, after about 9.8 h at an RH of 51%, which was around the third exchange of the saturated salt solutions. Although there was a lack of recorded videos, as explained above, the described stagnation of the cumulative video event count was considered valid, as the curve flattened significantly and remained constant at a very low level between hour 9.8 and hour 10.6. The corresponding video of the onset of the stagnation of the cumulative AE hit energy showed that the water of the droplet had already evaporated. Moreover, small pits had formed on the corresponding surface of the specimen. However, although the surface of the specimen seemed to be dry, small bubbling was still observed, which was also detected by the video processing algorithm. This bubbling took place in and in the near vicinity of already existing and emerging pits. This observation demonstrates that a seemingly dried-out surface does not mean that corrosion is no longer taking place. The electrolyte can remain in the pits and allow corrosion to progress further. This was expected for two reasons: (i) the drying of the pits simply takes longer, e.g., corrosion products around the pits build a barrier to the environment, or (ii), due to the deposits and surface conditions of the specimen, the equilibrium RH is reduced inside the pits [5,6]. At hour 23.5 when the RH reached its preliminary minimum of 33% and started to force the RH to rise again, the corresponding video showed a dry specimen and no more bubbles could be seen in the pits (cf. Figure 5b). About 20 min later at an RH of about 44%, again, small bubbling at the pits was observed, which led to a slightly progressive increase in the cumulative video event count until hour 51, when the RH increased to about 74%. At this point, small droplets could be seen on the surface of the specimen. In addition, the large NaCl crystals began to deliquesce due to the high RH that was very close to its equilibrium or deliquescence relative humidity (DRH) (cf. Table 1). From the point where the NaCl crystals started to deliquesce and thus become a liquid solution again (see Figure 5c), the curve of the cumulative video event count again increased significantly, as the liquid electrolyte promoted corrosion. Thus, more and probably larger bubbles formed that were detected by the video processing algorithm. The progress of corrosion was further observed by the increasing size of several pits (cf. Figure 5b–d). The curve of the cumulative AE hit energy started to increase again at hour 55 at an RH of 79%. The final and maximum RH was about 85%, which was reached at hour 76.7. Thereafter, the RH was rapidly decreased to a low level using silica gel. After a further 40 min, the cumulative AE hit energy clearly stopped increasing. The same held for the cumulative video event count. Again, about 40 min later, the RH reduction stagnation of the cumulative video event count was observed. Thus, due to the rapid reduction in the RH, the AE and video activity stopped simultaneously, i.e., the electrolyte in the pits also evaporated rapidly. These findings clearly confirm that corrosion took place during the period of significant AE and bubble activity and vice versa.The curve of the cumulative AE hit energy showed its steepest increase at the beginning and clearly flattened after 3.7 h. At hour 55, the curve started to increase with a slope similar to that observed between hour 4 and hour 7, i.e., during the evaporation of the droplet. These similar slopes seem plausible since similar conditions prevailed at these times (similar amount of highly concentrated electrolyte at similar locations). Thereafter, the curve gradually increased again and between hour 67 and hour 69, a slope that was almost as steep as at the beginning was observed. From hour 70 onward, the curve began to flatten. After the last forced increase in the RH, a small intermediate increase in the AE hit energy was observed at hour 75. The forced drying at hour 76.6 led to the final stagnation of the curve. This trend clearly shows that the most energy was generated during the phases of high RH, i.e., at the beginning when the droplet was still present and during the phase of sufficient deliquescence of the NaCl crystals, thus suggesting the correlation of the AE activity with the amount of electrolyte present on the specimen. This is also illustrated in Figure 8b by the scatter plot of the peak amplitude versus the AE hit energy. The four colored clusters represent the AE signals of the two phases when a large amount of electrolyte was present on the specimen (p1 and p3) and the two phases when no or a small amount of electrolyte was present on the specimen (p2 and p4) during the experiment (cf. annotations in Figure 8a). During phases p1 and p3, signals with high energies and peak amplitudes were detected, leading to the above-described steep increases in the cumulative AE hit energy. However, during phases p2 and p4, AE signals were also detected. These signals were less frequent and had significantly lower energies and peak amplitudes, which describes the stagnation of the cumulative AE hit energy during these two phases, as seen in Figure 8a.The slopes of the cumulative video event counts in the time spans at the beginning (hour 0 to hour 10) and at the start of the deliquescence of the NaCl crystals (hour 51 to hour 59) were similar. In the first time span, the focal plane of the camera was set near the droplet surface, where bubbles in the droplet are clearly visible. In the second time span, the focal plane was set at the specimen’s surface, where small bubbles in and near the pits were observed. From hour 59 onward, more and more electrolyte in the form of single droplets were present on the specimen due to the ongoing deliquescence of the NaCl crystals. However, due to the focal plane settings of the camera, bubbles appeared to be blurred in the droplets of the electrolyte and were less well detected by the algorithm. This is considered the reason for the flattening of the cumulative video event count curve between hour 59 and hour 72.5. From hour 72.5 to hour 76, the setting for the focal plane was changed (see the gray marked area in Figure 8a). The focal plane was moved slightly above the specimen’s surface. This modification made the bubbles partly visible. However, blurred bubbles were still present on the edges of the droplets, which were not detected by the algorithm. Thus, if there were several small droplets on the specimen, only a portion of the bubbles within these droplets was detected, i.e., changing the focal plane did not significantly change the trend in the cumulative video event count (cf. Figure 8a).From a global perspective, both cumulative curves showed similar trends that were both related to the measured RH. A stopping of the corrosion when the RH reached low values was observed in the videos (no pit growth or formation of new pits) and reflected in the cumulative video event count (no bubble activity). In addition, a re-onset of corrosion when the RH was increased again was observed. It was shown that the stopping and restarting of atmospheric corrosion could also be detected using the AE method, however, with some time delays with respect to the RH and bubble activity. It was observed that in the phases where just a small amount of solution was present on the specimen or only in the pits, fewer and smaller bubbles formed. It was believed that these small bubbles were too weak to generate AE events that exceeded the noise level and thus were not detected by the simple amplitude threshold approach used. The visual observation by the videos strongly suggests that bubbles were the main source of the measured AE signals, as very few events that were not bubbles were observed or detected. These other detected events were, e.g., the small but sudden growth of the small droplets during rewetting or the merging of these droplets during rewetting, and also the sudden shrinkage of the droplets during the phase of rapid RH reduction at the end of the experiment. Besides the presented comparison of the cumulative AE hit energy and cumulative video event count, a specific assignment of the bubble activity to the AE hits and vice versa was also attempted. Therefore, the synchronization of the corresponding two timelines was required, which was enabled by the flash signal of the camera (triggered at each video frame), which was connected to a parameter channel of the AE measurement system (cf. Figure 3). However, no clear assignment could be performed. The reasons for this were assumed to be the lack of depth of field; a too-low video framerate of 35 fps, i.e., a frame interval of approximately 29 ms compared to a typical AE hit duration of 5 ms; and the unclear effects of the bubbles (formation, detachment, movement, arrival at droplet surface, growth, bursting) that generated the AE event. Thus, the assignment of the bubble activity to the AE hits would require a refined test setup. 3.2. Frequency Analysis of Atmospheric Corrosion AE DataThe frequency analysis of the AE data of the atmospheric corrosion experiment revealed further AE activity that was not detected by the simple amplitude threshold approach. Therefore, a short-time Fourier transform (STFT) of the complete AE data, which also included the time periods where only noise without hits was measured, was conducted. Due to the very long duration of the entire AE measurement signal (approximately 4 days) compared to the duration of the individual AE hits (approximately 5 ms), a visual representation in the form of a spectrogram is not meaningful, and the dominant frequencies were instead plotted. The results of the STFT Z(t,f) were divided into n time segments, each of a Δt = 0.1 s duration. For each segment, the frequency f^i of the maximum of the absolute value of Z(t,f), i.e.,

f^i = arg maxfZ(t,f)t ∈ iΔt, (i + 1)Δt

(7)

with i = 0 … n − 1, was extracted. The STFT was calculated using the command signal.stft in the Python package scipy. The Hann window with a window length of w = 2500 samples and window overlap of w/2 was used, leading to a time resolution of 62.5 μs and a frequency resolution of 8 kHz. Figure 9 shows the results of this frequency analysis, together with the curve of the RH.Clear differences in the calculated frequencies between the wet and dry phases, i.e., the phases of high and low RH, can be seen. In particular, at the beginning, when the initially deposited droplet was present on the specimen, frequencies up to 1.5 MHz were yielded. The reduction in the RH, and thus the electrolyte on the specimen’s surface, also resulted in a reduction in the frequency. Between hour 12 and hour 48, the same frequencies were consistently observed. In this time span, a low RH prevailed and the NaCl crystals had not yet started to deliquesce. A detailed analysis of these frequencies (using the experimental measurement of the electromechanical impedance spectrum of the specimen with the applied PWAS; see [42] for this measurement method) revealed that some of them (e.g., 24 kHz, 40 kHz, 64 kHz, 88 kHz) fit to the natural frequencies of the specimen (e.g., 25.8 kHz, 42.3 kHz, 63.7 kHz, 85.9 kHz). Moreover, it was observed that the day (7 am to 8 pm) and night times (8 pm to 7 am) clearly influenced these frequencies (cf. detail view in Figure 9). It was assumed that this was related to the building technology of the laboratory (heating control, electrical infrastructure, lighting, etc.). However, the specific reason could not be identified. This influence had to be considered when attempting to determine the transition from frequencies indicative of corrosion activity to frequencies due to background noise. The transition indicating the stopping of corrosion due to the lowering of the RH was suggested to have occurred at approximately hour 7.5, which was between the stopping points determined by the cumulative AE hit energy and cumulative video event count. The onset of corrosion activity due to the increase in the RH was determined to have occurred at approximately hour 51.2, which matched the point of the cumulative video event count.The frequency behavior after hour 51.2 was again clearly different from the previous phase with a low RH. However, the maximum frequencies reached were lower than those in the initial phase where the full droplet was present on the specimen. The rapid reduction in the RH at the end of the experiment led to a short but significant increase in the frequencies, with peak values of nearly 2 MHz. A detailed view of these results revealed that this happened about 30 min after the start of drying out at an RH of 37% and ended about 12 min later, which fits well with the end of the increases in the above-mentioned cumulative quantities. The corresponding videos showed that at the beginning of this time span, the specimen surface was already dry but the remaining NaCl was still slightly soaked with some water. In the following 10 min, the last of the water evaporated, leaving dry NaCl deposits on the specimen’s surface, as can be seen in Figure 5e. Thus, it was expected that this high-frequency AE activity also originated from sources other than corrosion, e.g., the very fast drying of the surface and its deposits. However, due to the very short durations of these possibly disturbing additional AE sources, their effects on the experimental findings of the cumulative AE hit energy were excluded.

The frequency analysis shows that further information can be extracted from the AE data. The low bubble activity in the pits could not be fully detected with this frequency domain analysis. However, the transitions of the dominant frequency activities attributed to corrosion to the frequencies of the background noise and vice versa shifted the AE measurements closer to the corresponding observation points of the cumulative video event count.

3.4. Discussion of the AE Signal SourceTo systematically determine the source of the measured AE signals, all possible effects that may have provoked AE activity during the experiments are discussed. Figure 12a presents the classification of these effects. It is shown that atmospheric corrosion can be triggered on AA2024-T351 by a small droplet of a corrosive NaCl solution, as the visual observation clearly identified pits after some time. Hydrogen that formed during corrosion led to bubble activity, which was also seen in the videos, and the AE monitoring measured significant increases in the cumulative AE energy. All three observed phenomena, i.e., corrosion, bubbles, and increase in AE energy, were associated with the RH (cf. Figure 5 and Figure 8a), strongly suggesting that bubble activity (origination, detachment, movement, arrival at droplet surface, growth, bursting) was the main source of the measured AE signals, which has been repeatedly claimed in the literature for different kinds of corrosion tests [26,27,28]. However, it cannot be excluded that there were other unknown effects that generated measurable AE signals. These other effects may or may not have been related to corrosion (cf. Figure 12). Since we attempted to answer the question of whether atmospheric corrosion can be clearly detected by AE, sources not related to corrosion should ideally be excluded or known, considering that Figure 8 and Figure 10 show that the detected AE signals were dominated by corrosion-related effects. These results show that a sufficient amount of electrolyte (NaCl solution) results in corrosion and a large number of AE signals with comparatively high energies and peak amplitudes. In contrast, the different conditions with just a very small amount of electrolyte or Milli-Q water on the surface resulted in no or very little corrosion (cf. Figure 6 and Figure 11) and at the same time, generated only a small number of AE signals with significantly lower energies and peak amplitudes.However, the dominance of corrosion-related effects in the AE signals can be supported by considering Figure 12b. Figure 12b presents the classification of all possible effects that may have provoked AE activity during the experiments, but reduces the representation of the AE activity and its triggering effects, which are suppressed by the use of Milli-Q water instead of NaCl solution. The Milli-Q water generated corrosion and also measurable AE signals. However, as observed above, the corrosion caused by the Milli-Q water was found to be significantly less, which meant less formation of H2 (cf. Equation (1)), i.e., less bubble activity and also fewer other effects that triggered AE activity related to corrosion. Thus, it can be concluded that the measured AE signals were strongly dominated by corrosion-related effects. However, non-corrosion-related effects that contributed to the measured AE activity cannot be fully excluded but were shown to be of minor importance.There are various effects that have the potential to be sources of AE. These effects include (i) AE sources unrelated to corrosion. Environmental noise and interference were uncommon, as the experiments were conducted under laboratory conditions. Nevertheless, electric interference, which typically produces spike-like AE signals, was filtered out by not considering hits with counts less than five. AE sources caused by the effects of the used saturated salt solutions and the silica gel can also be excluded, as the AE data did not show sudden changes due to the exchanging of the salt solutions or the silica gel. (ii) AE sources related to corrosion but not directly related to bubble activity. A possible AE source could be the initiation or growth of microcracks (starting from pits) in the aluminum matrix caused by the formation of H2 bubbles from individual H+ ions or residual stresses in the matrix. However, microcracks can be excluded since they would also trigger measurable AE activity, and small H2 bubbles were observed (i.e., ongoing corrosion) only in some pits (cf. Figure 8 hour 7 to hour 10, and hour 51 to hour 55). Moreover, no significant residual stresses were expected in the simple, small, and thin specimens. A further effect triggering AE activity may be the dissolution into Al3+ ions in the pits (cf. Figure 1). However, this is a continuous process on an atomic level, i.e., low energy for AE, and thus it is also very unlikely to be a potential AE source. To fully identify the occurring corrosion mechanisms, further AE signal analysis by more advanced signal features [19,23] or post-mortem examination of the corrosion site by, e.g., SEM, is required. However, this is a matter for future research. (iii) AE sources specifically related to bubble activity. As presented in Figure 1, the main phases of a bubble are its formation, detachment from the metal surface, movement, arrival at the droplet surface, growth, and bursting. The sudden formation of gaseous H2 from single solid-state H+ ions could potentially generate a pressure wave leading to AE. This can be excluded since no AE hits were detected during the low RH phase when no droplets were present but small bubbles (indicating ongoing corrosion) were observed in some pits (cf. Figure 8 hour 7 to hour 10, and hour 51 to hour 55). Other potential sources are the detachment of bubbles from the specimen and subsequent bubble movement, arrival of bubbles at the electrolyte surface, and bubble growth, e.g., by the merging of the bubbles and the eventual bursting of bubbles (cf. Figure 1). All these effects must also occur when only a small amount of electrolyte is present in the pits; however, no AE hits were detected in this case (cf. Figure 8 hour 7 to hour 10, and hour 51 to hour 55). Therefore, it is assumed that the amount of electrolyte at the corrosion site influences whether or not AE hits can be detected. From the recorded videos, it was observed that a larger amount of electrolyte, i.e., larger droplets, leads to larger bubbles, which are expected to generate AE hits with larger amplitudes and energies. Moreover, a larger amount of electrolyte might create a higher counter pressure against the bubbles, resulting in bubble activity with higher energy, i.e., AE hits with higher amplitudes.

Thus, the results strongly suggest that bubble activity is the main source of the measured AE signals. Bubble bursting is assumed to be the most emissive source, but also the merging of the bubbles is expected to produce measurable AE signals. Consequently, the present monitoring approach may potentially be applied to other metals where gas bubbles form due to (atmospheric) corrosion, provided a sufficient amount of electrolyte is present.