La percepción de la variedad argentina – un análisis vertical.

Figure 5.2: Periodogram of example acoustic waveform collected by the ESD in operational underground conditions

trigger. For instance, if an indication of an on-going event can be seen in the initial period of the waveform, then an assessment needs to be made as to whether the rest of the waveform does in fact indicate a sounding event recorded as opposed to another anomalous cause. This assists in the data cleaning process in the next step of the research methodology.

Most of the acoustic energy generated from the event can be seen in the waveform peak from 13 ms to approximately 80 ms in Figure 5.1. The expected energy peak of the impact event occurs soon after the triggered start of the event, and then subsides back to ambient noise levels after 100 ms. This expectation of the waveform’s drop-off rate will also indicate during analysis whether a recorded noise is in fact from a rock being struck, as opposed to an environmental or anomalous noise that triggered the ESD.

5.3

Experimental set-up

The ESD is designed to be used in underground conditions and trained by experts on in situ examples of loose and solid rocks. A limitation on this research is that the rock examples that are demonstrated through this method by experts are not always proven actually to be structurally solid or loose, due to the fact that each rock mass is not subsequently barred down to test it. The examples are therefore expert assessments, but not objective fact. To mitigate this limitation in the initial development of the classifying algorithms, it is useful to have a controlled physical testing environment where the structural

states of the objects are fully known. This section describes the experimental test rig designed for the purposes of this research.

The intention of the test rig was to simulate a solidly attached rock and a loosely attached rock. To emulate this in a controlled environment a design of two separate blocks attached in various states of structural cohesion to a concrete roof was implemented. The schematic of the design can be seen in Figure 5.3. Component A is the design for the solidly attached block. This block is a solid marble cut-out that is solidly attached to the concrete roof by four iron bolts inserted at each corner of the block. Component B is the design for the loosely attached block. This block is also marble, but fractured in the middle and hanging from four iron bolts with a gap of 3 mm from the concrete roof.

Figure 5.3: Experimental rig layout showing (A) the solid block, and (B) the loose and fractured block

A photo of Component A is shown in in Figure 5.4, and a photo of Compo- nent B in Figure 5.5. Note the indicators on the blocks marking the locations where the blocks are to be impacted. This is to ensure that most of the pos- sible variations in the acoustic waveforms generated by the different areas of each block are consistently represented in the collected data set.

The aim of this experimental test rig was to enable the collection of an acoustic dataset that approximates the theoretical ideal dataset for this re- search. The attributes of the ideal acoustic dataset for this research that can be collected by means of the test rig must be as follows:

5.3. EXPERIMENTAL SET-UP 59

Figure 5.4: Experimental rig solid block

Fully descriptive: Each acoustic sample is accurately matched to a com- pletely known structural state. There is no ambiguity in the accuracy of the state.

Distinguishable: Acoustic samples generated from the solid block should be distinguishable from acoustic samples generated by the loosely attached block. A clear hyperplane should exist between the two subsets of the data collected.

Repeatable: The size of the sample base should be easily expandable. Access to the experimental rig should be easy on an ad hoc basis if more samples are required.

Noise-free: The samples should be realistically noise free. During the process of collection, background noise sources are monitored and avoided. The acoustic waveforms recorded by the ESD are purely from the impact of the sounding bar on the marble blocks at predetermined areas on the marble blocks.

During the collection of acoustic samples from the test set it is noted that the samples from the two blocks sound very distinct to a human observer. This clear disparity in the acoustic profile from the two sets of samples will make it easier to classify waveforms by means of a classification system. Caution is expressed that this approach therefore does not lead to ambiguous samples, and that this characteristic of the test platform disqualifies it from being a direct proxy of an underground mining working area.

A total of 200 acoustic samples evenly split between the solid and loose configuration of blocks is collected as the control sample for this research. These samples are intended to be used as demonstrators of distinguishable data separation and feature correlation in Chapter 6, and as basic validators of potential models in Chapter 7.