Evolving Neural Conditional Random Fields for drilling report classification

Identification of human errors and influencing factors: A machine learning approach

“…The present research has focused on previous studies that have used machine-learning strategies to classify textual narratives into safety and risk features. The sample also focused in industries with similar level of organisational and technological complexity as found in MATA-D, as well as those that have investigated at least one human factor as one of the features, such as aviation (Robinson et al, 2015), railway (Heidarysafa et al, 2018;Hughes et al, 2017), oil & gas (Ribeiro et al, 2020), civil construction (Goh and Ubeynarayana, 2017) and maritime industries (Grech et al, 2002). A comprehensive review of the application of machine-learning techniques in occupational accident analysis, however, mixing many industries with lower level of complexity is provided in (Sarkar and Maiti, 2020).…”

“…Previous studies have not classified full accident reports into a human reliability taxonomy -nor any attempts have been identified to expand databases of human reliability with the support of machine-learning, or within multiple industry sectors. For instance, only one specific human factor (situation awareness) has been analysed in maritime accident reports (Grech et al, 2002) while often the aim was to analyse near-misses or close call reports (daily basis reports that consist of small narratives from workers (Hughes et al, 2017), to support safety managers on having timely decisions upon risk controls (Goh and Ubeynarayana, 2017;Heidarysafa et al, 2018;Ribeiro et al, 2020;Robinson et al, 2015).…”

“…To minimise the randomised sample effect many studies present the metrics by variable (or sets of variables) instead by overall indicators (Goh and Ubeynarayana, 2017;Grech et al, 2002;Heidarysafa et al, 2018;Zhang et al, 2019). The difference in performance metrics can be more transparently depicted by error estimates (Ribeiro et al, 2020). The need for smaller uncertainties between estimates can also define the size of training and testing sets.…”

Linear Chain Conditional Random Field for Operating Mode Identification and Multimode Process Monitoring

“… 29 After that, Fang et al design a switching CRF based on multiple linear chain CRFs for multimode fault diagnosis. 30 As a powerful discriminative probabilistic model, a CRF is also used for working condition classification in oil drilling 31 and sucker rod pumps. 32 Similar to other machine learning algorithms, CRFs can also be directly used for modeling.…”