IoT SaaS - why automotive industry should care, and which AWS IoT or Azure IoT is better to use as a base platform for connected vehicle development

 The automotive industry is changing right before our very eyes. Today, services based on the CASE model are looming on the horizon. They are capturing an increasing market share and gaining more and more each year in total dollar value. What's in store for the automotive sector and how automotive enterprises can seize these opportunities?

New mobility services are emerging rapidly

By 2030, over 30 percent of the projected increase in vehicle sales due to urbanization and macroeconomic growth will be unlikely to happen owing to the shared mobility expansion.

In China, the European Union, and the United States, which are countries supporting  shared mobility solutions , the mobility market  could reach 28 percent annual growth from 2015 to 2030 . Of course- this would be the most optimistic scenario. FutureBridge specialists expect the shared mobility market to grow significantly over the next five to seven years at a CAGR of 16 percent from 2018, reaching  180 billion dollars by 2025 . How can the growing demand for new mobility services be explained?

On the one hand,  the automotive industry deals with  changing consumer preferences . One travels by car covering shorter distances, but much more frequently. And it doesn’t have to be by car at all, as new means of transportation are becoming more accessible.

On the other hand, soaring car prices (though cars lose their value a few months after the purchase) prompt us to search for other, cheaper alternatives that provide optimal driving comfort anyway.

How will companies relying on the traditional car ownership model respond to this trend? They will provide new services such as substitution models, in which, for a once-off monthly payment, you can have a new car with insurance, maintenance, roadside assistance, etc.  Subscriptions will soon account for about 15% of new car sales and should have risen to 25% by 2025. In this context, new mobility in the form of rental and ride-sharing services, which are also part of the transformation on the roads, also becomes significant.

The third thing is growing technology,  based on the CASE model(Connectivity, Autonomous driving, Shared mobility, Electrification,) that empowers the development of new mobility services on an unprecedented scale. According to Microsoft experts, by 2030 virtually all new cars will have been connected devices, functioning as data centers on wheels.

6 leading new mobility services

Carsharing

A short-term car rental model that allows users to choose a vehicle and pick-up/drop-off location. Users can determine vehicles and flexible rent times. Operators gain high ROI with high utilization and minimal staffing.

 Examples: citybee, E-VAI, fetch

Ride-hailing

A form of cab rental in which the drivers are usually contractors using their private vehicles rather than direct employees. The user has immediate availability and payment is handled through the operator. The benefits are also the ability to track and monitor journeys. For operators instead, traditional fleet costs must be handled by the drivers. It’s an easily scalable service.

 Examples: Uber, Lyft, Bolt, marcel, OLA

P2P Sharing

this service allows vehicle owners to rent their vehicles when they are not currently in use. BMW-run ReachNow is piloting a version of this type of service, which allows Mini owners to offer their currently unused vehicles for rent. The benefits for users are the lower costs than traditional vehicle rental. Meanwhile, the operator has no fleet to manage and gets access to an easily scalable model of business.

 Examples: HoppyGo, SnappCar

Carpooling

Allows users to join an already scheduled trip. The operating company acts as an "intermediary" through which rides can be announced and joined. Carpooling can apply both to people taking a trip alone and to those who want to share rides to reduce the total cost of the trip for a single passenger. It’s a cheap and environmentally friendly service. What is more, the operator has a higher margin per ride and no fleet to manage.

 Examples: BlaBlaCar, GoMore, liftshare

Car rental

The evolution of the traditional car rental by the day, allowing users to rent cars for different periods without the traditional hassle associated with this type of service. From the user's point of view, such new services enable an easier and quicker process of vehicle rental. Also, it’s possible to choose a vehicle before finalizing the rental. In turn, the operator has less staffing than a traditional rental and can utilize already existing fleets.

 Examples: Audi Silvercar, Hertz, Sixt, PORSCHE DRIVE, UBEEQO

Multimodal

An integrator of public transport mobility services, as well as other modes of transportation, such as public transportation, rail networks, and even cabs. The goal of such services is to get people from their starting point to their destination in the fastest, cheapest, or most efficient way, depending on individual needs. In this model, the operator gets access to additional potential users and has relatively low costs of deployment due to a lack of physical assets.

 Examples: FREE2MOVE, whim, Google Maps

Which new mobility services are growing the fastest?

Of the 55 providers of the aforementioned new mobility services operating in European countries, the most popular are those in the area of  carsharing  (51%) . The second most popular are  car rental services  (20%) , followed by  P2P sharing  (13%) .

In terms of ownership, most new mobility services were OEM owned (over 36%), although many of them were independent (over 38%). Also included were OEM invested services (31%).

Technologies and functionalities fueling the development of new mobility services

Mobility services are based on advanced software that uses, at least, the Internet of Things, to transfer data from the vehicle to the cloud. Then the individual information is available on the user's mobile application.

For services based on unmanned vehicle rental, modern security features have been considered when it comes to opening and closing the car.

With a view to minimizing possible problems, the developers of digital new mobility services are also introducing a  fault reporting option.

 Below is a selection of the most common functionalities and technologies in detail for each new mobility service in Europe.

All of these and other options provide guidance and a certain pattern of behavior for future developing OEMs.

Key factors crucial for the development of new mobility services

CASE trends provide new opportunities for the vehicles of the future. However, the interrelationships between software, in-car sensors, and electronic systems require a  huge amount of resources , especially when we are talking about reliable operations that translate into a competitive advantage of new mobility services and popularity among potential users.

 Therefore, if you want to develop in this area, consider at least these few factors.

1.     Cybersecurity    . In addition to creating huge amounts of code, what also matters is that your user data tracking processes comply with the standards and regulations that apply in your geographic region.

2.  Careful listening to user needs. In order to compete with technology start-ups, OEMs should focus on innovative digital solutions oriented towards     actual consumer expectations    . What matters is flexibility, when it comes to the portfolio of functionalities.

3.  Certainly,     emotion    is a factor that must be taken into account. Solution providers should care about providing unique experiences and sensations that will make the user eager to re-use a particular service, and in the process, spread it to their community.

4.     Flexibility and scalability    . You need to be prepared not only to meet the changing expectations of customers who come in with feedback but also to expand functionality to include those that competitors already have (or to offer completely innovative solutions).

5.     Being ready to expand the offering    . For example, with new types of vehicles: not only internal combustion but also hybrid, electric; not only cars but also city scooters, etc.

If you want to deal with the challenges that come with developing new mobility services and are considering the above and other growth factors, contact Grape Up. We can help you expand your business in terms of features and values appreciated by today's conscious consumers.

Machine Learning combined with edge computing gains a lot of interest in industries leveraging AI at scale - healthcare, automotive, or insurance. The proliferation of use cases such as autonomous driving or augmented reality, requiring low latency, real-time response to operating correctly, made distributed data processing a tempting solution. Computation offloading to edge IoT devices makes the distributed cloud systems smaller - and in this case, smaller is cheaper. That’s the first most obvious benefit of moving machine learning from the cloud to edge devices.

 Why is this article worth reading? See what we provide here:

•     Explaining why regular ML training flow might not be enough.  
•     Presenting the idea behind federated learning.  
•     Describing the advantages and risks associated with this technology.  
•     Introducing technical architecture of a similar solution.  

How can federated learning be used in the automotive industry?

Using  the automotive industry as an example, modern cars already contain the edge device with processors capable of making complex computations. All ADAS (Advanced Driver Assistance Systems) and autonomous driving calculations happen on-board and require rather significant compute power. Detecting obstacles, road lanes, other vehicles, or road signs happens right now using onboard vehicle systems. That’s why collaboration with companies like  Nvidia becomes crucial for OEMs, as the need for better onboard SoCs does not stop.

Even though the prediction happens in the vehicle, the model is trained and prepared using regular, complex, and costly training systems built on-premises or in the cloud.  The training data grows bigger and bigger making the training process computationally expensive, slower, and requiring significant storage, especially if incremental learning is not used. The updated model may take time to be passed to the vehicle, and storing the user driving patterns, or even images from the onboard camera, requires both user consent and adherence to local law regulations.

The possible solution for that problem is using a local dataset from each vehicle as small, distributed training sets and training the model in the form of “federated learning”, where the local model is trained using smaller data batches and then aggregated into a singular global model. This is both more computational and memory efficient.

What are the benefits of federated learning?

One of the important concepts highly associated with machine learning at edge is building Federated Learning on top of edge ML. The combination of federated learning and edge computing gives important, measurable advantages:

•  Reduced training time - edge devices calculate simultaneously which improves velocity compared to a monolithic system.
•  Reduced inference time - compared to the cloud, at the edge inference results are calculated immediately.
•  Collaborative learning - instead of single, huge training dataset learning happens simultaneously using smaller datasets - which makes it both easier and more accurate enabling bigger training sets.
•  Always up-to-date model in vehicle - the new model is propagated to the vehicle after validation which makes the learning process of the network automatic.
•  Exceptional privacy - the omnipresent problem of secure channels for passing sensitive user data, anonymization, and storing personal user data for training purposes is now gone. The learning happens on local data in the edge device, and the data never leaves the vehicle. The weights which are being shared cannot be used to identify the user or even his driving patterns.
•  Lack of single point of failure - the data loss of the training set is not a threat.

Benefits from these concepts contain both cost savings and accuracy improved, visible as an overall better user experience when using the vehicle systems. As autonomous driving and ADAS systems are critical, better model accuracy is also directly associated with better security. For example, if the system can identify pedestrians on the road in front of vehicles with accuracy higher by  10%, it can mean that an additional 10% of collisions with pedestrians can be avoided. That is a measurable and important difference.

Of course, the solution does not come only with benefits. There are certain risks that have to be taken into account when deciding to transition to federated learning. The main one is that compared to the regular training mechanisms, federated learning is based on heterogeneous training data - disconnected datasets stored on edge devices. This means the global model accuracy is hard to control, as the global model is derived based on local models and changes dynamically.

This can be solved by building a hybrid solution, where part of the model is built using safe, predefined data, and it is gradually enhanced by federated learning. This brings both worlds closer together - amounts of data impossible to handle by a singular training system and stable model based on a verified training set.

Architectural overview

To build this kind of system, we need to start with the overall architecture. Key assumptions are that the infrastructure is capable of running distributed, microservices-based systems and has queueing and load balancing capabilities. Edge devices have some kind of storage, sensors, and SoC with CPU, and GPU capable of  training the ML model .

Let’s split it into multiple subsystems and consider them one by one:

1.  Swarm of connected vehicle edge devices, each one with connected sensors and ability to recalculate model gradient (weights.)
2.  Connection medium, in this case fast, 5G network available in the car
3.  Cloud connector, being a secure, globally available public API where each of the vehicle IoT edge devices connect to.
4.     Kubernetes cluster    with federated learning system split into multiple scalable microservices:

a) Gradient verification / Firewall - system rejecting the gradient that looks counterfeit - either manipulated by 3rd party or being based on fictional data.
b) Model aggregator - system merging the new weights into the existing model and creating an updated model.
c) Result verification automated test system - system verifying the new model on a predefined dataset with known predictions to score the model compared to the original.
d) Propagating queue connected to (S)OTA - automatic or triggered by user propagation of updated model in the form of an over-the-air update to the vehicle.

A firewall?

The firewall here, inside the system, is not a mistake. It is not guarding the network against attacks. It is guarding the model against being altered by cyber attacks.

Security is a very important aspect of AI, especially when the model can be altered by unverified data from the outside. There are multiple known attack vectors:

•  Byzantine attack - regarding the situation, when some of the edge devices are compromised and uploading wrong weights. In our case, it is unlikely for the attacker to be omniscient (to know the data of all participants), so the uploaded weights are either randomized but plausible, like generated Gaussian noise, or flip-bit of result calculation. The goal is to make the model unpredictable.
•  Model Poisoning - this attack is similar to the byzantine attack, but the goal is to inject the malicious model, which as a result alters the global model to misclassify objects. The dangerous example of such an attack is by injecting multiple fake vehicles into a model, which incorrectly identifies the trees as “stop” road signs. As a result, an autonomous car would not be able to operate correctly and stop near all trees as it would be a cross-section.
•  Data Poisoning - this attack is the hardest to avoid and easiest to execute, as it does not require a vehicle to be compromised. The sensor, for example, camera, is fed with a fake picture, which contains minor, but present changes - for example, a set of bright green pixels, like on the picture:

This can be a printed picture or even a sticker on a regular road sign. If the network learns to treat those four pixels as a “stop” sign. This can be painted, for example, on another vehicle and cause havoc on the road when an autonomous car encounters this pattern.

As we can see, those attacks are specific to distributed learning systems or machine learning in general. Taking this into account is critical, as the malicious model may be impossible to identify by looking at the weights or even prediction results if the way of attack was not determined.

There are multiple countermeasures that can be used to mitigate those attacks. Median or distance to the global model can be calculated and quickly identify rogue data. The other defense is to check the score of the global model after merging and revert the change if the score is significantly worse.

In both cases, the notification about the situation should be notified, both to operators as a metric and to a service that gives scores to the vehicle edge devices. If the device gets repeatedly flagged as wrong-doing, it should be kicked out of the network, and investigation is required to figure out if this is a cyberattack and who is the attacker.

Model aggregation and test

As we know, taking care of the cybersecurity threats specific to our use case, now the important step is merging the new weights with the global model.

There is no one best function or algorithm that can be used to aggregate the local models into global models by merging the individual results (weights). In general, very often average, or weighted average gives sufficient results to start with.

The Aggregation step is not final. The versioned model is then tested in the next step using the predefined data with automated verification. This is a crucial part of the system, preventing the most obvious faults - like the lane assist system stopping to recognize roadside lines.

If the model passes the test with a score at least as good as the current model (or predefined value), it’s being saved.

Over-the-air propagation

The last step of the pipeline is enqueueing the updated model to be propagated back to vehicles. This can be either an automatic process as in  continuous deployment directly to the car or may require a manual trigger if the system requires additional manual tests on the road.

A safe way of distributing the update is using the container image. The same image may be used for tests and then run in vehicles greatly reducing the chance of deploying failing updates. With this process, rollback is also simple as long as the device is able to store the previous version of the model.

The results

Moving from legacy, monolithic training method to federated learning gives promising results in both reduced overall system cost and improved accuracy. With quick expansion of 5G low-latency network and IoT edge devices into vehicles, this kind of system can move from theoretical discussions, scientific labs, and proofs of concepts to fully capable and robust production systems. The key part of building such a system is to consider the cybersecurity threats and crucial metrics like global model accuracy from the start.

 Ever since Henry Ford implemented the first production line and launched mass production of the Ford Model T, the automotive industry has been on the constant lookout for ways to boost performance. This aspect has become even more relevant today, given the constant market and social unrest. Coming to rescue supply chain management and product lifecycle optimization is predictive maintenance. Not only OEMs, but the entire automotive industry: insurers, car rental companies and vehicle owners are benefiting from the implementation of this technology.

Predictive maintenance explained

Predictive maintenance is an advanced maintenance approach that utilizes data science and predictive analytics to anticipate when equipment or machinery requires maintenance before it faces a breakdown.

The primary aim is to schedule maintenance at optimal times, considering convenience and cost-effectiveness while maximizing the equipment's longevity. By identifying potential issues before they become critical, predictive maintenance significantly reduces the likelihood of equipment breakdowns.

Various types of maintenance strategies are employed in different industries:

1.     Reactive Maintenance:    Also known as "run-to-failure," this method involves waiting for equipment to fail before conducting maintenance. Therefore, unscheduled downtime and higher repair costs may occur.
2.     Periodic Maintenance    : This approach entails performing maintenance tasks at regular intervals, regardless of the equipment's condition. It helps prevent unexpected breakdowns but may lead to unnecessary maintenance if done too frequently.
3.     Smart Maintenance    : Smart maintenance utilizes advanced technologies like IoT devices and sensors to monitor equipment in real-time and identify anomalies or potential failures.
4.     Condition-Based Maintenance    : This strategy relies on monitoring the equipment's condition while it is in operation. Maintenance is only carried out when data indicates a decline in performance or a deviation from normal parameters, optimizing maintenance schedules and reducing unnecessary work.
5.     Predictive Maintenance    : The most advanced type of maintenance uses real-time operational data and predictive analytics to forecast when maintenance is required. It aims to schedule maintenance before equipment failure occurs based on data-driven predictions, thus minimizing downtime, reducing costs, and prolonging equipment lifespan.

Predictive maintenance employs various techniques, such as vibration analysis, acoustic monitoring, infrared technology, oil analysis, and motor circuit analysis. These methods enable continuous equipment condition monitoring and early detection of potential failures, facilitating timely maintenance interventions.

Differentiation between predictive maintenance and preventive maintenance

 Predictive maintenance hinges on the real-time condition of assets and is implemented only when the need arises. Its purpose is to anticipate potential failures by monitoring assets while they are actively operational. Unlike  preventive maintenance , this approach is rooted in the current operational state of an asset rather than statistical analysis and predetermined schedules.

Essential steps in creating a predictive maintenance solution

Predictive maintenance solutions utilize a combination of sensors, artificial intelligence, and data science to optimize equipment maintenance.

The development of such solutions varies depending on equipment, environment, process, and organization, leading to diverse perspectives and technologies guiding their creation. However, there are steps common to every project: data collection and analysis, model development and deployment, as well as continuous improvement.

Here is a step-by-step process of how solutions are developed in the  automotive industry :

•     Data Collection    : Relevant data is collected from sensors, equipment logs, vehicle diagnostics, telemetry, and other sources. This data includes information about the performance, condition, and behavior of the vehicles, such as engine temperature, fuel consumption, mileage, and more. Telematics systems can provide real-time data on vehicle location, speed, and usage patterns, while maintenance logs record historical maintenance activities, repairs, and part replacements.

•     Data Preprocessing    : The collected data is organized, and prepared for analysis. Data preprocessing involves cleaning the data by removing outliers or erroneous values, handling missing values through imputation or interpolation, and converting the data into a suitable format for analysis.

•     Feature Engineering    : Important features or variables that can provide insights into the health and performance of the vehicles are selected from the collected data. These features can include engine vibration, temperature, fuel consumption, mileage, and more. Feature selection step involves identifying the most relevant features that have a strong correlation with the target variable (e.g., equipment failure). It helps to reduce the dimensionality of the data and improve the model's efficiency and interpretability. Later, selected features are transformed to make them more suitable for modelling. The process may include techniques such as logarithmic or exponential transformations, scaling, or encoding categorical variables.

•     Model Development    : Machine learning algorithms are applied to the selected features to develop predictive models. These models learn from historical data and identify patterns and relationships between various factors and equipment failures. The algorithms used can include regression, decision trees, random forests, neural networks, and more.

•     Model Training and Validation    : The developed models are trained using historical data and validated to ensure their accuracy and performance. This involves splitting the data into training and testing sets, evaluating the model's performance metrics, and fine-tuning the model if necessary.

•     Deployment and Monitoring    : The trained models are deployed into the predictive maintenance system, which continuously monitors real-time data from sensors and other sources. Telematics systems are used to collect GPS and vehicle-specific data, which it transmits through different methods (cellular network, satellite communication, 4G mobile data, GPRS) to the central server. The system detects anomalies, recognizes patterns, and provides insights into the health of the vehicles. It can alert maintenance teams when potential issues are detected.

•     Continuous Improvement    : The predictive maintenance solution is continuously improved by collecting feedback, monitoring its performance, and updating the models and algorithms as new data becomes available.

Most common problems in deploying predictive maintenance solutions

Implementing predictive maintenance solutions in a fleet of vehicles or in a vehicle factory is a process that requires time, consistency and prior testing. Among the main challenges of rolling out this technology, the following aspects in particular are noteworthy.

Data integration

Integrating data from many sources is a significant barrier to implementing predictive maintenance solutions. To accomplish this with a minimum delay and maximum security, it is necessary to streamline the transfer of data from machines to ERP systems. To collect, store, and analyze data from many sources, businesses must have the proper infrastructure in place.

Insufficient data

Lack of data is a major hindrance to implementing predictive maintenance systems. Large amounts of information are needed to develop reliable models for predictive maintenance. Inadequate information might result in inaccurate models, which in turn can cause costly consequences like premature equipment breakdowns or maintenance.

To get over this difficulty, businesses should collect plenty of data for use in developing reliable models. They should also check that the data is relevant to the monitored machinery and of high quality. Businesses can utilize digital twins, or digital representations of physical assets, to mimic the operation of machinery and collect data for use in predictive maintenance systems.

Process complexity

Transitioning from preventive to predictive maintenance is complex and time-intensive. It requires comprehensive steps beyond technology, including assembling a skilled team and managing upfront costs. Without qualified experts versed in software and process intricacies, project success is doubtful.

High costs

The implementation of predictive maintenance programs comes with substantial costs. These upfront expenses pose challenges, including the need to invest in specialized sensors for data collection, procure effective data analysis tools capable of managing complexity, and possibly hire or train personnel with technical expertise.

To address these hurdles, collaboration with specialized vendors and the utilization of cloud-based solutions can prove cost-effective. Additionally, digital twin technology offers a way to simulate equipment behavior and minimize reliance on physical sensors, potentially reducing overall expenses.

Privacy and security issues

The implementation of predictive maintenance involves extensive data collection and analysis, which can give rise to privacy concerns. Companies must adhere to applicable data protection laws and regulations, and establish proper protocols to safeguard the privacy of both customers and employees. Even though predictive maintenance data may be anonymized and not directly linked to specific individuals, it still necessitates robust security measures, since preventing data breaches and unauthorized access to vital company information is crucial for overall success.

What Are the Benefits of Predictive Maintenance?

Life cycle optimization, stock management, or even recycling management - in each of these fields predictive maintenance can bring substantial benefits. And this is not only for OEMs but also for fleet operators, transportation or logistics companies. And even for the end user.

Below we list the key benefits of implementing  predictive maintenance in an automotive-related company:

•     Extended lifespan:    Predictive maintenance technology detects early signs of wear and potential malfunctions in-vehicle components such as engines, transmissions, and brakes. By addressing these issues proactively, vehicles experience fewer major breakdowns and continue to operate efficiently over a longer period.

•     Cost savings:    By addressing issues at an early stage, automotive companies can avoid expensive breakdowns and prevent further damage. This proactive approach not only reduces the need for costly replacement parts but also minimizes the labor and operational costs associated with major repairs, resulting in significant long-term cost savings.

•     Minimized downtime    : Through continuous monitoring and analysis, predictive maintenance predicts when maintenance or repairs are needed and schedules them during planned downtime. This minimizes the likelihood of unexpected breakdowns that can disrupt operations and lead to extended periods of vehicle inactivity. By strategically timing maintenance activities, vehicles spend more time on the road.

•     Increased efficiency    : Any iissues are detected early, enabling timely corrective actions. This proactive approach leads to improved fuel economy, reduced emissions, and overall enhanced efficiency. Vehicles operate at their peak performance, contributing to a more sustainable and environmentally friendly fleet.

•     Enhanced security:    Constant monitoring for abnormal vibrations, temperature variations, and fluid leaks ensures that potential issues compromising vehicle safety and security are detected promptly. By addressing these concerns before they escalate, predictive maintenance contributes to ensuring the security of both the vehicle and its occupants. This feature is particularly valuable in critical applications where reliable vehicle performance is paramount, such as emergency response scenarios.

•     Avoiding over-maintenance    : If you over-maintain corporate resources, it can have the same negative consequences as when failing to maintain them on time. With predictive maintenance, you can focus on maintaining crucial resources at the best possible time and with the best possible results.

•     Compliance with required standards and regulations    : Laws and regulations related to vehicle production are constantly evolving and pushing OEMs to make numerous production changes (e.g. the legislation related to EV production). Predictive maintenance allows you to better suit the new expectations of legislators and monitor the points of production that are most dependent on the legal context.  
  
•     Easier management of parts and materials    : As connected cars diagnostic systems become more sophisticated, drivers have the option to make small repairs sooner and keep their vehicles in a better condition. All this means that OEMs and licensed repair shops need fewer parts and can better manage supply chains.

 Predictive maintenance clearly is not a one-size-fits-all solution for all sectors. Notably, it will work well for high production volumes and short lead times and anywhere you need to ensure reliability, security and convenience.

The automotive industry is a perfect fit for this model. As shown in the examples featured in the second part of the article, the top players in the market are tapping into this technology.

According to  Techsci Research , “  The global predictive maintenance market was valued at USD 4.270 billion in 2020 and is projected to grow around USD 22.429 billion by 2026”.