The first industrial revolution began in the late 18th century, with the mechanization of the textile industry.
Nearly a century later, in the second half of the 20th century, the third industrial revolution appeared with the emergence of computers and the beginnings of automation, when robots and machines began to replace human workers on the assembly lines.
And now, we are in the fourth industrial revolution called “Industry 4.0” – it is the next phase in the digitization of the manufacturing industry. Four factors drive Industry 4.0:
- Staggering rise in data volumes, computational power and connectivity
- Evolution of advanced analytics and machine learning capabilities
- Emergence of new technologies such as touch interfaces and augmented-reality systems that enable the interaction between humans and machines
- Improvements in transferring digital instructions to the physical world, such as 3-d printing and advanced robotics.
Industry 4.0 is rapidly shifting manufacturing from isolated, optimized cells of business processes, systems, and resources to fully integrated data and product flows across corporate borders.
Therefore, manufacturing companies have to build up capabilities in IoT service development and operation. Put differently, the achievement of “integrated production for integrated products”.
A lot of these manufacturing firms will find this difficult, because it is not in their DNA. It is also not about developing additional IT skills. To make the change possible, the value propositions of these companies should evolve, this means a change in almost all parts of the organization, from engineering to sales right from through to aftermarket services.
In this blog post, I will walk you through the two most widely discussed use cases in industry 4.0 – predictive maintenance and connected cars.
Predictive maintenance(PdM) techniques are designed to help determine the condition of an in-operation equipment/machine to predict when maintenance should be performed.
The main promise of predictive maintenance is to allow convenient scheduling of corrective maintenance, and to prevent unexpected machine failures.
You would have read many articles stating that predictive maintenance is the same or a part of condition-based maintenance. Based on our experience, we chose to call it out separately as the scope and implementations are quite different.
The condition-based maintenance primarily uses rules and anomaly detection techniques to predict an outcome. On the other hand, predictive maintenance analyzes volumes of historical data, trends, correlations, trend and machine specifications to anticipate an outcome.
I have written a separate blog post that explains how condition-based maintenance works in real-world.
Now, let’s dive into the use cases.
Predictive maintenance is the ability of the system to predict a machine failure.
Predictive maintenance(PdM) can reduce or even abolish unplanned downtime by predicting when a machine needs checkups or when a machine will become faulty.
Predictive maintenance is one of the most widely discussed use cases in the IoT ecosystem.
Interesting examples of predictive maintenance
A simple use case would be to predict the remaining life cycle of an asset and when maintenance is required using the streams of data we get from the assets such as the actual ‘wear and tear’ data of its parts. Imagine a dashboard, which lists down the assets and its metadata, like manufacturing date, installed date, type etc. along with its actual usage. It also depicts external factors and predictions on the remaining life cycle of the asset and a maintenance date. These factors can be used to plan minimum maintenance downtime, schedule spare parts delivery and ensure maintenance is executed with least impact.
Also, most manufacturers typically have historical maintenance records of the systems and usage dates in some form – this can be a valuable input to predict the maintenance activity.
Let’s take an example of a leading elevator manufacturing company that supplies elevators across the globe.
Can they use an IoT system to predict when their elevator lift cables should be replaced? Manufacturing innovations are happening in elevator cables, like using super-light carbon fibre ropes that increases the lifespan of the cables. Changing the lift cables is an expensive maintenance activity, and its failure can have significant downtime.
Factors such as availability of new lift cables, specialized technicians, and compliance check can have a considerable impact on the business operations.
To carry out any predictive maintenance for elevator lift cables, the manufacturer needs to decide the data points that would be required to predict the failure.
As part of its product design, the manufacturer had probably installed a sensor to track the running time or distance served by the cable, a sensor to detect if the elevator is descending faster than its designated speed, start and stop instances of the elevators. Sensor input together with the cable’s specified life expectancy can be used to predict when the lift cables need to be replaced. In an actual scenario, many more such datasets need to be provided to predict outcomes.
The role of machine learning in predictive maintenance
Predictive maintenance involves building out machine learning models based on volumes of data. Developing machine learning models require considerable time and effort. It’s virtually impossible to expect a system to devise a predictive model which is 100% accurate (not even humans operate with that level of accuracy :)), but should be considerable enough to suggest a cause of possible failure with reasonable accuracy.
Open-source scalable machine learning models like Spark MLlib or commercial offerings like SPSS from IBM or Azure ML for Microsoft can aid in building predictive models. The traditional predictive maintenance machine learning models like Support Vector Machines, Logistic Regression, and Decision Trees are based on feature engineering, which is manual construction of correct features( attributes) using domain expertise and similar methods. This actually makes it hard to reuse the model as feature engineering is specific to the problem scenario.
Using deep learning algorithms, we can automatically extract the right features from the data, eliminating the need for manual feature engineering. Among the deep learning networks, Long Short Term Memory(LSTM) networks are used in predictive maintenance domain since they are very good at analyzing huge volumes of time series sensor data.
The model once developed can be integrated into your IoT platform to predict outcomes in real time.
How companies are implementing predictive maintenance solutions
Deutsche Bahn and Siemens have launched a 12 month pilot project to provide data to support predictive maintenance of the Class 407 Velaro D high-speed train fleet.
Data received from the operational Velaro D fleets will be consistently analyzed at the Siemens mobility centre in Munich, supplementing diagnostic data available onboard. This will help the maintenance team to identify the impending faults and malfunctions as well as the sources of these problems early on. Specialist technicians will then recommend corrective actions to the technicians at the Deutsche Bahn’s workshop.
“The objective of the pilot project is to the precisely align maintenance work with the vehicle’s actual status. With intelligent algorithms and precise analytics, availability is increased,” said Jochen Eickholt, CEO of the Siemens Mobility Division.
Another good example of predictive maintenance is ThyssenKrupp. ThyssenKrupp partnered with CGI and Microsoft Azure to send alerts when their elevators need repairs. Predictive maintenance system sends alerts when an elevator is about to go out of function and even teaches the technicians the areas of error.
Before we deep dive into the connected car use case, let’s first start with a formal definition of a connected car. As per Wikipedia – “A connected car is a car that is equipped with internet access, and usually also with a wireless local area network. This allows the car to share internet access with other devices both inside as well as outside the vehicle.”
From a hardware perspective, the car can connect to the internet, by plugging in devices to the OBD2 port of the car to extract the vehicle data. Now a days, insurance companies are issuing dongles – devices that plug directly into the OBD2 port and connect wirelessly to a network – to customers as a way to achieve discounts. This generally involves using data pulled from the car’s OBD2 connection to analyze driving habits and award a discount for low-risk behaviour. Allstate’s Drivewise program, for example, looks at speed, how quickly the driver applies brakes, the number of miles driven, and when a person drives.
Interesting connected car use cases
From a software perspective, here are some connected car use cases:
- Location Tracking
- Real-time performance monitoring of the car
- Condition based maintenance
- Predictive maintenance
- Driver assistance
- Behaviour analysis of the driver
- Recommendation based on driving patterns
- Speak on
The above uses cases apply to both individuals and fleet management firms who could monitor cabs remotely, create high speed alerts, analyze driver’s behaviour pattern, Street-view( Google or similar), wait time and actual distance covered.
Overview of the connected car ecosystem and related services
The above use cases are developed and deployed as cloud services. Primarily, the OBD2 data and GPS location is made available continuously to the core IoT platform. Let’s inspect some of the information available as part of the OBD port. The values can be retrieved by using the PIDs (parameter Ids) codes from OBD2 port. For instance, to retrieve the engine speed, the standard PID 12 code needs to be used. Some of the information that can be retrieved are:
- Vehicle speed
- Engine RPM
- Coolant temperature
- Air flow rate
- Absolute throttle position
- Absolute load value
- Fuel status
- Fuel pressure
- Battery voltage
The above is only a minimal set of data from the car. Your car is already equipped with hundreds of sensors and with the car now being connected, the data generated through the car is being utilized for various other use cases.
Once the OBD and GPS data is available to the IoT core platform, the platform can start consuming it. The IoT core platform would read the continuous stream of vehicle data, uncompress the same, persist and analyze the data, execute the required cloud services (condition monitoring, custom alerts, behaviour analysis) and notify the user. A mobile or web-based application is provided to the user to view the data in real time offered by various services.
Given below are the set of use cases with brief implementation details:
Location Tracking: Provides a map view and location of the car on a mobile or web application.
Real-time performance monitoring of the car:Provides real-time graphical and tabular view of the performance data of the car from OBD port.
Condition based maintenance:We have discussed this earlier. The same approach is applicable for a vehicle, to track performance, detect anomaly (faults from normal deviations) and provide alerts if maintenance is required.
Predictive maintenance:This phase performs predictive maintenance activity of the vehicle using the various approaches discussed earlier.
For example, Mahindra Reva is an electric micro-car that comes with built-in connectivity, which provides customers, dealers, and operators with real-time insight into car status and performance.
This solution monitors the health of the electric vehicle and helps field support staff to identify the root cause of potential problems. It also enables customers to access information about the vehicle, as well as allowing remote access to certain parameters.
This feature provides various assistance to the driver, like route planning which can minimize fuel consumption or advising the user to not take the regular route due to weather conditions and suggest alternatives.
Driver behaviour analysis:
Classifies the driver’s ability over a period of time based on the vehicle data (vehicle speed, RPM, throttle position, etc.), idling, accelerometer (tracking rapid lane changes), brake pressure, accelerator pedal position, GPS data (GPS calculated vehicle speed) and driver past history.
The parameters can be fed to a machine learning model which can then classify the driver as – novice, unsafe, neutral, assertive or aggressive.
Recommendation based on driving patterns:
This involves analysing the driver’s driving patterns over a period and providing recommendations on how best to drive the car and utilize the capabilities of the car.
For instance, suggesting how to save fuel based on driving patterns (driving at second gear constantly in a geared car), less use of brakes and slowing down the accelerator during speed breakers or avoiding sharp turns and yielding at specific locations (tracked via GPS, accelerometer and steering angles).
You can converse with your car in natural language about its features, and it responds or provides recommendations based on the data collected from above use cases. Imagine this being an intelligent SIRI system that knows everything about your car and answers your questions effectively by understanding the context.