Rocky soil gives excavation and tunnel-boring machines a hard time. The machine’s expensive cutting tools quickly wear out or break. With neuro-fuzzy modelling, Dr.
Mario Alvarez Grima, from Cuba, is able to predict how quickly.
Predicting the amount of cutting tools needed for a project is important for companies when tending an offer. Huge amounts of money are involved and underestimating cutting-tool costs can result in a company losing millions of guilders.
The influence of certain types of rock properties and rock mass on the wear of cutting tools, however, is difficult to predict. Many of the rock’s charateristics are often unknown, as are relationships between different variables. Traditional mathematics, therefore, won’t do for making these calculations.
“About five years ago, a colleague, Dr. Verhoef, suggested trying fuzzy modelling techniques, which is a different kind of mathematics that most engineers aren%t yet used to,” says Alvarez Grima, who defended his PhD thesis on September 18th. “These techniques make use of the human tolerance for incompleteness, uncertainty, imprecision and vagueness in decision-making.”
Neuro-fuzzy methods attempt to mimic the way engineers reason. The models learn from data and allow for expressing the knowledge of a given process linguistically, such as ‘the permeability of the rock is high.’ Soft models also combine theoretical knowledge with laboratory and field observations, thereby putting scientific knowledge in the hands of industry, according to Alvarez Grima.
Soft computing, however, isn’t a magic tool, he warns. “You must be open with industry and tell them about the model’s weaknesses. Ideally, a model is based on physical and mechanical principles. If all the fundamental principles aren’t available, soft computing may help in constructing a workable and realistic model from measured data and theoretical knowledge. However, there are limitations. You must know some basic relations and you must have enough typical data to make a model.”
To collect the field data for his first model, Alvarez Grima studied machines at work in countries like Italy, French and Germany, for two and half years. “I followed the machine of the company that sponsored my research, collecting data about how fast cutting-tools needed to be replaced under various conditions and how fast the machine advanced in different geological settings.”
Spark
Industry engineers and practitioners speak a different language than geological scientists at universities, according to Alvarez Grima, who worked in Cuban industry for seven years prior to coming to Delft. Alvarez Grima’s computer program, therefore, asks questions in the language used by engineers. In the field, they have to fill in experimental data, such as what they see when they hit a rock: do you, for example, see a spark or does the material crumble when hit with a geological hammer? Another thing they have to estimate is the depth of the soft layer above a rocky layer. The scientific definitions that must be filled in are explained quite extensively if engineers click on the term in the computer program. The mathematics, however, are hidden: when using the computer program, you don%t see any formulas.
For his second project, Alvarez Grima didn%t carry out the observations himself. He was lucky and could use a database from MIT-Texas University, offering 640 tunnel-boring machines. “We can now predict how fast a tunnel-boring machine may penetrate a certain soil type. Our neuro-fuzzy method proved to be more accurate and easier to interpret for people in industry, compared to conventional statistical models.”
The use of neuro-fuzzy techniques is currently growing exponentially in geological sciences, according to Alvarez Grima. “For me, it was nice to start in a new research field. From the start of our research projects, editors of the scientific journals were very interested in our articles and asked us to write more.That gave us a lot of energy for continuing our investigations.”
Rocky soil gives excavation and tunnel-boring machines a hard time. The machine’s expensive cutting tools quickly wear out or break. With neuro-fuzzy modelling, Dr. Mario Alvarez Grima, from Cuba, is able to predict how quickly.
Predicting the amount of cutting tools needed for a project is important for companies when tending an offer. Huge amounts of money are involved and underestimating cutting-tool costs can result in a company losing millions of guilders.
The influence of certain types of rock properties and rock mass on the wear of cutting tools, however, is difficult to predict. Many of the rock’s charateristics are often unknown, as are relationships between different variables. Traditional mathematics, therefore, won’t do for making these calculations.
“About five years ago, a colleague, Dr. Verhoef, suggested trying fuzzy modelling techniques, which is a different kind of mathematics that most engineers aren%t yet used to,” says Alvarez Grima, who defended his PhD thesis on September 18th. “These techniques make use of the human tolerance for incompleteness, uncertainty, imprecision and vagueness in decision-making.”
Neuro-fuzzy methods attempt to mimic the way engineers reason. The models learn from data and allow for expressing the knowledge of a given process linguistically, such as ‘the permeability of the rock is high.’ Soft models also combine theoretical knowledge with laboratory and field observations, thereby putting scientific knowledge in the hands of industry, according to Alvarez Grima.
Soft computing, however, isn’t a magic tool, he warns. “You must be open with industry and tell them about the model’s weaknesses. Ideally, a model is based on physical and mechanical principles. If all the fundamental principles aren’t available, soft computing may help in constructing a workable and realistic model from measured data and theoretical knowledge. However, there are limitations. You must know some basic relations and you must have enough typical data to make a model.”
To collect the field data for his first model, Alvarez Grima studied machines at work in countries like Italy, French and Germany, for two and half years. “I followed the machine of the company that sponsored my research, collecting data about how fast cutting-tools needed to be replaced under various conditions and how fast the machine advanced in different geological settings.”
Spark
Industry engineers and practitioners speak a different language than geological scientists at universities, according to Alvarez Grima, who worked in Cuban industry for seven years prior to coming to Delft. Alvarez Grima’s computer program, therefore, asks questions in the language used by engineers. In the field, they have to fill in experimental data, such as what they see when they hit a rock: do you, for example, see a spark or does the material crumble when hit with a geological hammer? Another thing they have to estimate is the depth of the soft layer above a rocky layer. The scientific definitions that must be filled in are explained quite extensively if engineers click on the term in the computer program. The mathematics, however, are hidden: when using the computer program, you don%t see any formulas.
For his second project, Alvarez Grima didn%t carry out the observations himself. He was lucky and could use a database from MIT-Texas University, offering 640 tunnel-boring machines. “We can now predict how fast a tunnel-boring machine may penetrate a certain soil type. Our neuro-fuzzy method proved to be more accurate and easier to interpret for people in industry, compared to conventional statistical models.”
The use of neuro-fuzzy techniques is currently growing exponentially in geological sciences, according to Alvarez Grima. “For me, it was nice to start in a new research field. From the start of our research projects, editors of the scientific journals were very interested in our articles and asked us to write more.That gave us a lot of energy for continuing our investigations.”
Comments are closed.