Design and characterization of a smart bit for in-situ force measurement, using capacitive load cells and acoustic spectra analysis to determine rock type and tool wear in rock excavation process

Underground miners continue to be exposed to hazards on a routine basis. This is best mitigated by removing the operator from the hazardous locations while increasing overall productivity. This work investigates methods for determining tool wear and material type with a sensing system, which would enable operators to make decisions using objective feedback from a safer location. Continuous mining machine operators must determine tool wear and material type during operation, and they place themselves at risk when they get close to the cutting interface to do so. Vibration frequencies, acoustic emissions, and cutting forces are all known to vary with rock type and tool wear conditions, and this work captures those changes with varied cutting test conditions. Three different sensor designs were tested: a capacitive load cell with non-linear dynamics used to classify material type and tool wear conditions, an acoustic sensor used to classify tool wear, and a capacitive load cell with linear dynamics used to measure the cutting forces. Full scale rock cutting tests are performed using a linear cutting machine at the Earth Mechanics Institute at the Colorado School of Mines campus. Analytical models of the capacitive sensors are developed for this research, and they are shared to guide future designs. This work discusses the sensors’ different sensitivities to input force. These analytical models also guide the choice of classification methods used to determine material type and tool wear. The machine-learning classification methods perform well for the experimental conditions. The capacitive load cell with linear dynamics experienced forces up to 80 kN in magnitude while cutting coal samples cast in concrete. When the sensor data was used with a small neural network regression and a 2nd order polynomial expansion, it is able to measure rock cutting forces with a mean absolute error less than 4 kilonewtons and an R 2 score greater than 0.8 under tested conditions. This is slightly better than the linear regression performance on the same data, with a mean absolute error less than 6 kilonewtons and an R 2 score greater than 0.6. When the sensor data was used for material and tool wear classification, the Support-Vector machine using fast Fourier spectra magnitude of short duration samples of signal, around 0.2 seconds, performed the best. Other machine learning methods explored for classification applications in this research include multi-layer perceptron neural networks and k-nearest neighbors classifiers, which are tested using the acoustic sensor. The objective differences in frequency spectra of the forces on the tool are shown in each of the journal articles for each sensor and application. These differences in frequency spectra can enable new sensor applications for improving mining safety and efficiency by enabling remote operation from a greater distance.

Name thesis:degree_name

Doctor of Philosophy (Ph.D.)

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor dc:publisher

Colorado School of Mines. Arthur Lakes Library

Year dc:date.issued

2024

Author dc:creator

• Oltmanns, Austin F.

Advisors dc:contributor.advisor

• Petruska, Andrew J.
• Petrella, Anthony J.

Committee members dc:contributor.committeemember

• Rostami, Jamal
• Nilsen, Oyvind
• Han, Qi

• capacitive sensor
• force sensor
• Fourier transform
• machine learning
• support vector machine
• wear classification

Statement dc:rights

• Copyright of the original work is retained by the author.

Language dc:language.iso

eng, English

Dc Identifier Other

T 9835

OAI identifier oai:identifier

oai:repository.mines.edu:11124/180183