Development of a Modality-Invariant Multi-Layer Perceptron to Predict Operational Events in Motor-Manual Willow Felling Operations

Abstract

Motor-manual operations are commonly implemented in the traditional and short rotation forestry. Deep knowledge of their performance is needed for various strategic, tactical and operational decisions that rely on large amounts of data. To overcome the limitations of traditional analytical methods, Artificial Intelligence (AI) has been lately used to deal with various types of signals and problems to be solved. However, the reliability of AI models depends largely on the quality of the signals and on the sensing modalities used. Multimodal sensing was found to be suitable in developing AI models able to learn time and location-related data dependencies. For many reasons, such as the uncertainty of preserving the sensing location and the inter- and intra-variability of operational conditions and work behavior, the approach is particularly useful for monitoring motor-manual operations. The main aim of this study was to check if the use of acceleration data sensed at two locations on a brush cutter could provide a robust AI model characterized by invariance to data sensing location. As such, a Multi-Layer Perceptron (MLP) with backpropagation was developed and used to learn and classify operational events from bimodally-collected acceleration data. The data needed for training and testing was collected in the central part of Romania. Data collection modalities were treated by fusion in the training dataset, then four single-modality testing datasets were used to check the performance of the model on a binary classification problem. Fine tuning of the regularization parameters (α term) has led to acceptable testing and generalization errors of the model measured as the binary cross-entropy (log loss). Irrespective of the hyperparameters’ tunning strategy, the classification accuracy (CA) was found to be very high, in many cases approaching 100%. However, the best models were those characterized by α set at 0.0001 and 0.1, for which the CA in the test datasets ranged from 99.1% to 99.9% and from 99.5% to 99.9%, respectively. Hence, data fusion in the training set was found to be a good strategy to build a robust model, able to deal with data collected by single modalities. As such, the developed MLP model not only removes the problem of sensor placement in such applications, but also automatically classifies the events in the time domain, enabling the integration of data collection, handling and analysis in a simple less resource-demanding workflow, and making it a feasible alternative to the traditional approach to the problem.

Improving the Event-Based Classification Accuracy in Pit-Drilling Operations: An Application by Neural Networks and Median Filtering of the Acceleration Input Signal Data

“…The same classification performance metrics and errors were computed in the testing phase, where the Log loss was used as a metric to evaluate the generalization errors. Following the testing phase, the probabilities of given data points to fall within a given class (“Drilling”, “Other”, “Stopped”) were extracted and plotted against the magnitude data of those datasets, following the procedure described in [ 13 ]. The criteria used for evaluating the training and testing models and for selecting the best alternative were the classification accuracy (CA) and the Log loss error.…”

“…At this level, manual-dominated tasks have been approached in forestry under the umbrella of the so-called human activity recognition, which has been implemented by the use of various data collection platforms and machine learning techniques e.g., [ 11 , 12 ]. A similar approach has been used to monitor tool or machine-supported tasks, at least when such machines were not equipped with built-in production monitoring systems [ 8 , 9 , 10 , 13 ]; this approach still justifies its relevance due to the low to intermediary mechanization level of forest operations that still prevails in many parts of the world [ 14 , 15 , 16 ].…”

“…The fact that a human cannot perform the tasks of a machine in a natural way makes AI a complement to global development. It is likely, therefore, that artificial intelligence will soon be present in every aspect of our life since it can be already found in a wide range of industries, including medicine [ 20 , 21 , 22 , 23 ], communication [ 24 , 25 , 26 , 27 ], marketing [ 28 , 29 , 30 ], agriculture [ 31 , 32 , 33 ] and, of course, forestry [ 8 , 9 , 10 , 11 , 12 , 13 , 34 , 35 ]. In relation to operational monitoring by the means of automated time studies, recent research on the topic in forestry have proven that high classification accuracies may be achieved by the use as inputs in the machine-learning algorithms of raw signals outputted by various type of sensors such as accelerometers, gyroscopes and sound-pressure level meters [ 8 , 9 , 10 , 11 , 12 , 13 ].…”

“…It is likely, therefore, that artificial intelligence will soon be present in every aspect of our life since it can be already found in a wide range of industries, including medicine [ 20 , 21 , 22 , 23 ], communication [ 24 , 25 , 26 , 27 ], marketing [ 28 , 29 , 30 ], agriculture [ 31 , 32 , 33 ] and, of course, forestry [ 8 , 9 , 10 , 11 , 12 , 13 , 34 , 35 ]. In relation to operational monitoring by the means of automated time studies, recent research on the topic in forestry have proven that high classification accuracies may be achieved by the use as inputs in the machine-learning algorithms of raw signals outputted by various type of sensors such as accelerometers, gyroscopes and sound-pressure level meters [ 8 , 9 , 10 , 11 , 12 , 13 ]. In addition, signals outputted by accelerometers coupled with ML techniques have proven very useful not only in the forestry but also in other engineering disciplines such as those dealing with infrastructure and its monitoring [ 36 , 37 , 38 ].…”

“…While the approach of using the raw data as inputs may prove to be useful for fully automated, real-time applications, because it may ease the computational effort, in many ways its use in a processed form for offline modeling and improvement is important for the science. In addition, there are hints that processing of the raw signals and parameter tunning in machine learning applications may be useful to improve the outcomes in terms of classification accuracy; in particular, the use of median filters, tuning the model size and its learning rates could be among the good strategies to increase the classification performance [ 8 , 9 , 10 , 11 , 13 ].…”