Manual P-Phase Picking on Noisy Acoustic Emission/Microseismic Data

"The acoustic emission/microseismic monitoring (AE/MS) technique has widely been used in the mining environment for the monitoring of rock mass stability. A successful implementation of the method is hinged on being able to accurately determine the P- or S-wave arrival which is then used in the determination of the source of an event. Determination of the arrival phase is usually performed visually by a human expert. However, the mining environment is characterized by high noise levels and different noise types of varying intensities. Therefore, performing manual phase picking on time series data acquired from the AE/MS monitoring system can sometimes be tedious and time consuming if not impossible. In this paper, a proposal for improving manual P-wave phase arrival picks is discussed. The method involves the filtering of the AE/MS data using a stationary discrete wavelet transform (SDWT) method. After filtering, appropriate scales/levels of the multi-level decomposition process are then selected for further processing. The first step in the picking process after filtering, requires the computation of the signal power and the root mean square (RMS) using the signal amplitudes at each scale. The preliminary P-phase arrival time is picked using the results of the calculated power and RMS values respectively. The calculated RMS values are then superimposed on the calculated power values to determine the final arrival picks. If the arrival time picks obtained by both methods coincide or are within 5 to 10 sample points, the pick is considered reliable. The accuracy of the method was verified using AE/MS data from two datasets obtained from two separate underground mines. The results of the study showed that the picking method is reliable and robust. Based on the data studied, it can be concluded that the method will help reduce the time usually spent in processing AE/MS data from mines. INTRODUCTION A successful identification and location of the source of acoustic emission/microseismic (AE/MS) events is mainly dependent on the onset time of the P-wave phase. In many seismic networks, initial picks are performed by picking algorithms followed by manual picks by human experts. If the waveform signal-to-noise ratio (SNR) is low, the P-wave onset time picks by the human expert is generally reliable and accurate [1]. However, for waveforms with high SNR, picking of the P-wave onset could be a major challenge. Also, due to the volume of data acquired daily, manual processing could be stressful and time consuming [2]. These challenges continue to provide motivation for alternative means of providing reliable and efficient ways of improving manual phase picking. The AE/MS technique has been an intergral part of safety monitoring in mines [3] due to its capacity to identify zones of instability in a monitoring area. To obtain an accurate source of an AE/MS event however depends on the quality of the data used for P-wave phase picking [4, 5]. AE/MS data from mines are generally contaminated with excessive background noise generated by the different mining operations. The AE/MS signals of interest are mostly buried in background noise which makes it difficult to pick the P-wave phase arrivals [4]. Processing AE/MS data acquired in noisy environment remains a major challenge due to the weak nature of AE/MS signals [6]. As noted by [4] most of the data recorded in the mine environment are mainly due to events not related to AE/MS events of interest. The complexity of data in this type of environment makes the detection and picking of P-wave arrival by human experts a very difficult task."