The Hype and the reality
One of the important applications of IoT in manufacturing and operations that we keep hearing about is predictive maintenance. When planned, modeled and executed properly, it is an extremely powerful capability to build. If you have ever work on manufacturing floors, you are probably pretty familiar with the impact of the impact machine breakdowns have on manufacturing operations.
According to a research from Gartner, only 15% of predictive maintenance initiatives succeed in the first execution attempt. So the reality is, even though theoretically the logic seems simple, proper planning, structuring and execution is extremely important. In this post, we will discuss some of these aspects as they pertain to predictive maintenance initiatives.
The IoT data flow for maintenance: An example
Let us assume that we are an external consulting company and we have been retained by an Australian mining company for operations consulting work. One of the aspects that they are struggling with is unexpected breakdown of their critical mining equipment. They already have an architecture in place, with sensors integrated with the mining equipment, a bucket wheel excavator (BWE) that is used in strip mining (see the picture below).
A typical bucket wheel excavator uptime is in the range of 40-60% so there is always a significant opportunity to improve the uptime. The current model that the company is using is preventive or reactive. They also already have sensors installed on the equipment that relays the data back. If something breaks, it is fixed, otherwise, maintenance is performed on a pre-defined schedule or, in rare instances, when data from sensors indicate an anomaly.
Defining a value proposition
First, we need to define what is the value proposition of leveraging IoT in this scenario. The value proposition here is to increase uptime and eliminate unscheduled downtimes.
Defining a model
To define a model, you need to understand what are the key aspects that need to be monitored, as it pertains to predicting failures. Now since we have smart consultants in our hypothetical consulting company who have a mechanical engineering background, they have identified the following three parameters, that will be dependent variables in our model:
- Stress and strain
The consultants then determine that these parameters are functions of other variables and capture them as shown below:
- Temperature = f(load, angular velocity, vibration frequency)
- Friction = f(torque, temperature)
- Stress/Strain (Remember Young’s modulus?) = f(force 1, force 2, …….force n)
Note that sensors that are already installed on our BWE will help us capture the variables like load, angular velocity etc.
Application of the model: Proactive and Predictive maintenance
Data generated by sensors can be used to take proactive measures to make sure that the three key parameters above always remain within a reliable range. For example, if the joint temperature is between 68 and 230 degrees Fahrenheit, the probability of failure is temperature related failures are minimum but if it over the maximum range limit, an actuator can actuate water jets to proactively cool the joint temperature.
This is the nirvana application of data from sensors- leveraging the models formulated above and historical failures data to predict future failures. This can be realized with predictive analytics, which tries to recognize a failure signature in the data of the part in question. If recognized, it will predict that the part, let us say the joint in our example, will fail. We do need a long enough history of the same part to have a statistically relevant cause and effect model.
Summary: The value proposition chart for smart maintenance
Objective: Maintain assets better
- Temp= f(load, angular velocity, vibration frequency)
- Friction = f(torque, temperature)
- Stress/strain = f(force 1, force 2,…..force n)
- Use rules engine to highlight if variables are outside norms
- Remotely actuate cooling or lubrication
- Remotely modify operating software to limit lift
- Predict failures by interpreting model variables over time
- Prescribe change to be actuated in the product to avoid future failures
Views my own