PROJECT BACKGROUND
One of the major global challenges of the modern era is combating climate change. The most commonly accepted method of doing this is by reducing carbon emissions, the primary contributor to the greenhouse effect. An approach to achieving this is by integrating more wind power into the energy supply. However, public opinion on wind turbines has pushed wind generation increasingly from onshore to offshore, which has presented new and challenging engineering problems.
​
Offshore wind turbines are subjected to various aerodynamic and hydrodynamic forces, leading to the structures experiencing a significant amount of fatigue. Due to the nature of the location of the turbines, the remaining fatigue life of the structures can be difficult to estimate because of the positioning of the sensors. A commonly applied method to try and quantify the amount of fatigue within the turbines is Structural Health Monitoring (SHM). However, there are still many issues associated with this method. These issues include the unreliability of strain gauges and the costs associated with the quantity of strain gauges that would be required to monitor an entire structure. Strain can only be measured at the location of the strain gauge, hence why so many would be needed to cover a full turbine.
​
This project intends to address the issues associated with SHM by applying a technique known as Virtual Sensing. Virtual sensing allows a strain to be estimated at any point on a structure using sensors located elsewhere on the structure. For this project, accelerometers will be used to estimate the strain. There are various types of virtual sensing; for this project, virtual sensing using a modal expansion approach in conjunction with operational modal analysis (OMA) was used.
​
For the project, a simple cantilever structure was developed and accelerometers were initially placed equidistant from each other. The accelerometers recorded data when the cantilever was subjected to vibration. Once the data was obtained, OMA was utilised to determine modal parameters, mode shapes and natural frequencies. These parameters were then used to update a finite element model (FEM) of the cantilever beam, which was repeated until the FEM parameters matched the ones obtained experimentally. This allowed the strain to estimated with a high degree of accuracy. Applying this technique could allow the remaining fatigue life to be determined.
​
PROJECT AIMS
Demonstrate the potential of the Virtual Sensing technique using modal expansion and OMA on a simple structure using accelerometers and a strain gauge for verification.
Demonstrate that this technique is possible for structural health monitoring of structures where fatigue life cannot be accurately determined due to the inability to position sensors
Estimate strain at a point on a simple aluminium cantilever beam, where no sensors are located
Design and build a functioning test rig that can be reused for future projects
Develop a recommendation into the feasibility of virtual sensing for SHM
​
​
THE TEAM
Members and their Roles
Callum Roberts - Team Leader and Experimental Work
Calvin Nisbet - Technical Lead and Mathematically Modelling
Cameron Murphy - FEA and Simulation Assistant
David Garcia Cava - Project Supervisor
Paul Sweeney - Experimental Lead
Sophie Isbister - FEA and Simulation Lead
​
The team also had an industry partner from Brüel & Kjær, Dmitri Tcherniak. The team would also like to express their gratitude to the time Dmitri dedicated to helping us.
PROJECT MANAGEMENT
Project management was an important component of this group project and was one key to the success of the project. How the project management was to be carried out was outlined in the Statement of Purpose and Contract, which is available in the downloads section.
The interim report outlines how the project was effectively managed throughout the first semester, this can also be in the downloads section. The final report gives an in depth analysis of how the group felt the project was managed and gives recommendations on areas that could have been improved.
​
Below is a list of activities and procedures that were put in place by the team to ensure good project management:
Each member of the team completed a personality test to determine what role they would be most suited to.
The Stage Gate Method* was implemented.
A Gantt Chart was used to plan activities and was regularly reviewed and revised.
Weekly reports were published to keep all members of the team well informed of progress being made and meetings that occurred. The use of weekly reports was designed to mitigate the risk of the project falling behind.
​​
* The Stage Gate Method is a system developed for project management, the premise of which is that each stage of the project has a gate that needs to be passed for the project to move onto the next stage. At the end of each stage it is reviewed and the team as a whole decide if the project should progress or if more work is needed on that stage. For this project a very basic form of Stage Gate was implemented, which worked on a "go or no go" basis.
Up-to Date Progress
PROJECT PROGRESS
Statement of Purpose and Contract - completed 13th October 2017
Initial Literature Review - completed November 2017
Cantilever Design - completed October 2017
Test Rig Design - completed November 2017
FEM Model Generated - completed October 2017
Interim Report - completed 17th November 2017
Oral Presentation 1 - completed 12th January 2018
Risk Assessment - completed 8th January 2018
Cantilever Manufacture - completed 9th February 2018
Test Rig Manufacture - completed 9th February 2018
MATLAB Script Written - completed January 2018
Practice Testing - completed 7th February 2018
Presentation to industry partners Brüel & Kjær - completed 9th February 2018
OMA testing - started 13th February 2018 - completed 22nd February 2018
Strain Gauge built - completed 19th February 2018
Final Report Literature Review - completed 24th February 2018
Strain estimation testing - started 26th February 2018 - completed 13th March 2018
Compiling Final Report - started 23rd February 2018
MATLAB Code Updated - completed 3rd March 2018
Simple Cantilever
TEST SPECIMEN
When designing the test specimen a set of design criteria were identified. These included: ​
The beam must be a simple geometry, i.e., a rectangular with constant cross-section.
The cantilever must vibrate with one dominant set of modes, i.e., flap, edgewise or torsional. Flap was the preferred mode.
The cantilever must be made from low cost metal, aluminium was chosen for this.
The first natural frequency of the cantilever should be between 5Hz and 10Hz. MATLAB was used to verify this once dimensions of the beam had been decided.
The first torsional, longitudinal and transverse vibration mode frequency should be greater than the first 4 flap frequencies. Again, MATLAB was used to verify this.
For the MATLAB verification free-fixed boundary conditions were used. The fixed end was assumed an ideal support, with no damping, rotational or linear displacement.
The image above shows the manufactured test specimen. The dimensions of the test specimen are (750x50x30)mm.
In addition to the MATLAB verification, an FEM model was generated in ANSYS Workbench. This gave the same theoretical natural frequencies as MATLAB. The ANSYS model was to be used primarily for OMA.
ANSYS MODEL
The finite element analysis (FEA) carried out during this project was a relatively simple modal analysis of a cantilever beam using Ansys Workbench 17.1. This FEA was key to several areas of the project, such as beam design, OMA and strain prediction.
Firstly, the geometry of the physical cantilever beam, used in the experiment, was determined, refer to test specimen section, and a model of this was generated in ANSYS. A modal analysis was performed to acquire the natural frequencies for all mode shapes. To reaffirm the first natural frequency of the cantilever had to be between 5Hz and 10Hz.
​
The initial analysis was conducted with ANSYS' standard aluminium material properties and boundary conditions in order to compare with the first few practice experiments. This was then later adapted to more accurately match the main experiment for this project. The objective of OMA was to increase the accuracy of the finite element model to match the experimental data produced. See OMA section for the methodology.
Finally, the FEA was used to accurately predict the strain at a specific location on the beam where no accelerometer is placed. This was achieved by updating the model through OMA and then comparing the strains from the FEA, the experiment and the results obtained from MATLAB. When all strains matched from these three techniques, then it was concluded that the finite element model was accurately predicting the strains at any location along the beam.
​
The final geometry of the beam was 700x50x3mm. This geometry met the natural frequency requirements and a trial and error method was used until these requirements were met. However, this geometry was slightly different to the actual manufactured geometry. This was because the rig was manufactured with 50x25mm steel beams so the manufactured geometry had to be an additional 50mm. This was to ensure there was still 700mm of material free to vibrate and 50mm of clamped material.
​
TEST RIG
Design and Manufacture
Similarly to the cantilever design, the first step of designing the test rig was identifying design criteria. The criteria consisted of:
The rig must hold the test specimen veritcally.
The rig must hold the test specimen with no possible movement at the base, laterally and rotationally.
The rig must be self-supporting and stable under its own weight.
The rig must be able to hold a shaker, required to excite the test specimen.
The rig must be reusable, so that an independent can repeat the experiments.
​
Rig Design:
Once the design criteria had been identified the rig was designed to the specifications using CAD software. There were three components of the rig that were critical to ensuring the criteria was met. These were: the base, the shelf for holding the shaker and how the specimen would be held (clamping).
The base was designed as a horizontal square section with major dimensions of (600x600)mm. The base was made from steel and was welded together at the corners so that it was stable and self-supporting.
The shelf was added into the design to ensure the shaker could be held by the rig. To do this, two additional steel beams were welded to the base in a vertical orientation. Holes were then drilled into these beams so that the height of the shelf could be adjusted, allowing the specimen to be excited in various locations. The shelf itself was a beam of steel that was bolted horizontally between the two vertical beams. A platform was placed in the middle of the shelf for the shaker to sit on.
The final component was the clamping of the specimen. For this, the specimen was clamped in place by two pieces of steel. One piece was welded across the middle of the base and the second bit was left free. Holes were drilled in both pieces, which allowed the specimen to bolted in place and held with no movement at the base. Since one of the sections of steel was adjustable made it easier to take the test specimen in and out the rig.
The images below show the CAD design of the rig and the manufactured rig respectively.
​
EQUIPMENT
Brüel & Kjær supplied the team with equipment and software to use for the experimental work. This included:
Frontend Type 3560 D (top left image)
Power Supply for Frontend
Power Amplifier Type 2706 (top right image)
Mini Shaker Type 4810 - with rig (middle left image)
Force Transducer Type 2312 - with plastic stringer and nuts
Charge Converter Type 2624 (with low noise cable AO 0038-D-012)
Accelerometers Type 4507 B004 and Type 4507 B005 (middle right image)
Accelerometer plastic mounts
All the cables required to connect the equipment
In adddition, a strain gauge was built for more information see the Strain Gauge Section
STRAIN GAUGE
For our project a traditional strain gauge systems was not sufficient for the measurement of strain in this case, as traditional methods rely on the balancing of a Wheatstone bridge, however the balancing of a Wheatstone bridge in a dynamic strain problem is not possible. As such the unbalanced potential difference across the bridge must be converted into a strain signal.
​
The potential difference across the bridge was found by applying electrical theory and Kirchoff's voltage and current laws. The gauge factor was then used to obtain a time dependent strain, this derivation can be downloaded below.
​
A small bread board-circuit has been created to perform this task, and is jacked into the B&K front end by a BNC connection. Note: LED’s serve a protective purpose dissipating part of the voltage as B&K front end is limited to 5V.
​
For practical reasons, the breadboard layout is slightly more complex than the basic circuit diagram suggests.
For redundancy in circuitry, two independent strain gauge circuits were ‘built’ into a single rail energised board such that if a component failed on one circuit the other would be usable to continue the experiment.
Guard resistors were also added to limit the potential peak supply across the bridge and therefore not interact with the voltage limit of the B&K equipment. LED’s were also added at the head of the circuit (which also have a voltage reducing effect) however more importantly indicate when the circuit is energised.
​
Purchased components were all 12Ω resistors with a component tolerance of 1% and a set operating temperature in excess of 80°C. As mentioned above in order to simplify equations, the strain gauge was also of basic resistive value 12Ω, and this significantly simplified equations as well as naturally balancing the measurement circuits.
​
The strain gauge was attached to the specimen using a thin layer of adhesive and tape to keep the strain gauge flat to the specimen.
​
Please click on the images to enlarge and look through the gallery.
TESTING
An Overview
Once the specimen was secured in the test rig, the sensors could be attached and testing carried out. This set-up can be seen in the figure on the left. All the electronic equipment for testing was supplied by Brüel & Kjær for which the group were very grateful.Â
Initially, the accelerometers were placed on the test specimen with beeswax, equidistant apart. These accelerometers were connected to the frontend, which was in turn connected to a laptop where the data was recorded using PULSEÂ Labshop.Â
The first stage of testing was operational modal analysis (OMA). These tests would give modal parameters of the specimen, in particular the mode shapes of the first four natural frequencies. The theoretical mode shapes were obtained from the FEM model. These tests were repeated until the mode shapes of the model and the specimen were near identical (major diagonal MAC Numbers of at least 0.95).Â
After OMA had been carried out, the rig was set-up for strain estimations, which would be verified using a strain gauge. This part of the experiment demonstrated if virtual sensing was a valid technique to obtain strain anywhere on the specimen. MATLAB was used to obtain the strains.
OPERATIONAL MODAL ANALYSIS
Testing - Stage 1
The initial stage of testing was operational modal analysis (OMA). For this PULSE LabShop, MATLAB and OMA software were required.
The accelerometers were attached to the cantilever using beeswax. For the first part of OMA, LabShop was used to record the vibration signal. Recording occurred for approximately 10 seconds. From this, the lowest and highest natural frequency, given the spatial resolution due to the number of accelerometers, can be obtained. The period of the lowest natural frequency was then calculated and multiplied by 200 to obtain the minimum recording length required to find the lowest mode shape using OMA. For the test specimen, this was found to be 45 seconds. The highest obtainable frequency was multiplied by 3 to determine the minimum value of the sampling frequency, which was found to be approximately 1.5 kHz.Â
Following this, recording for mode shapes was then conducted. This was done by hitting the specimen at different points at random intervals throughout the recording, this can be seen in the video below.Â
The files obtained were then converted to MATLAB files. In MATLAB, a script was written to convert the data into a .binary file to make the data compatible with the OMA software. In addition, a .cfg file had to be written in Notepad, which would allow the OMA software to obtain the mode shapes and create animations of them. The mode shapes were obtained using frequency domain decomposition (FDD). Finally, the mode shape vectors were obtained by converting the data from the OMA software into .uff files.
The theoretical mode shapes were then obtained from ANSYS, which allowed the MAC number to calculated. More information on MAC numbers can be found in the literature review download. To summarise, if the major diagonal MAC numbers were less than 0.95 and the off-diagonal MAC numbers greater than 0.2 then the accelerometers had to be moved and the FEM updated. The experiment was repeated until the major diagonal MAC numbers were at least 0.95.
Four iterations of operational modal analysis had to be completed to obtain the desired MAC numbers and these can be found in the results section.
STRAIN ESTIMATION
Testing - Stage 2
After operational modal analysis had been successfully carried out, the next stage of testing could begin. This was using virtual sensing to estimate the strain at a point on the specimen, where a sensor had not been placed. For verification, a strain gauge was attached to the specimen where the team wanted to estimate the strain.Â
Prior to testing, the beam had to be prepared. This involved gluing plastic clips onto the beam for the accelerometers in the locations determined from OMA. The accelerometers were then placed in the plastic clips. Next the strain gauge had to be attached to the beam. In addition, a fifth accelerometer was attached using beeswax on the opposite side of the specimen from where the strain gauge. This was done to ensure OMA had been carried out successfully, as predicting the acceleration at a random point on the cantilever would be simpler than the strain.Â
Once the cantilever had been prepared it was fixed back in the test rig for the testing to begin. On the laptop connected to the equipment, LabShop was opened. The two new sensors were added to the template, which was then activated and the shaker started (setting the waveform to random and the frequency span to 400Hz). Once all the sensors were detecting a signal, the recorder was started. Data was recorded for 5 minutes per test. The data was then exported to MATLAB files using TimeData Recorder. The data was then inputted into MATLAB and the virtual strain and acceleration determined. This was then compared to the actual recorded strain and acceleration in the virtual location.
LIVE STREAM
Here, you can find a live stream to the lab showing the current round of testing.
CONTACT
James Weir BuildingÂ
University of Strathclyde
75 Montrose St
GlasgowÂ
G1 1XJ
+44 779 448 6799