Artemy Malkov, PhD
July 27, 2022
NLP in manufacturing

Dmitry Malkov, Yuliya Rubtsova, Alexey Veselovsky, Marina Molchanova, Sergey Berezin and Artemy Malkov

In this article, we’ll be describing specific cases for manufacturing and near-manufacturing processes. Cases are grouped into categories with an indication of the technology stack from NVIDIA with which they could be developed. Now, let’s look at the cases.

Conversational AI

Technologies: NVIDIA Riva, Triton Server, TensorRT, CUDA

NVIDIA Riva speech skills
NVIDIA Riva speech skills

Give a voice to IoT (Internet of Things)

Challenge: Manufacturing of any kind has a certain degree of risk associated with it. Any delay in getting the information about critical condition changes in manufacturing facilities may result in a significant threat to workers’ health and wellbeing — not to mention huge monetary losses.

Solution: If a manufacturer has sensors or cameras integrated with NLP, the responsible specialists can be notified immediately in the event of any risk. IoT-enabled sensors or cameras can detect anomalies, and integration with NLP can allow people in charge to be informed without delay via text or voice. In this case, necessary actions may be initiated ASAP to mitigate the potential damage.

Interaction with customers

Challenge: Have a fast reaction after receiving a wide range of support queries, complaints, and service requests.

Solution: Your company should maintain a strong relationship with your customers, vendors, and dealers by resolving their requests and questions. Smart conversational AI can automate almost 90% of support services for you. NLP and ML help bots figure out problems or requests faster, and lets them respond with relevant information or action. It also reduces the workload of your employees and can capture leads on your website and from other channels. All collected leads can be collected in your database or CRM. That way, your sales team can take them towards conversion faster.

Interaction with vendors

Challenge: Companies invest a lot of time in communicating with multiple vendors, understanding their services, and negotiating deals. Before choosing a vendor, companies have to carry out research, which can consume a tremendous amount of time.

Solution: Chatbots can take up this mundane task, helping companies make quicker and fairer deals without involving their workforce. AI-enabled chatbots can get in touch with vendors, interact with them to understand their services, and make a decision on whether or not the vendor is right for their company. Besides this, chatbots, trained with human negotiation abilities, can bargain and close deals at a better price. While chatbots negotiate with vendors for the best possible deals, employees have more time to take up strategic roles.

Enter supply data

Challenge: Supply data is unstructured.

Solution: To optimize supply chains by collecting necessary information, chatbots request information from suppliers through adaptive interviews.

Coordination with other departments within the company

Challenge: Workers should get all the information they require without any delay.

Solution: All the necessary information on every process can be provided by chatbots. Such AI chatbots could be integrated with the company’s ERPs or other legacy systems to pull out all the required information. Questions could be asked in natural language. So, managers or employees obtain some details on manufacturing or shipping activities in real-time. That helps managers make more informed decisions about inventory or supplies.

Internal Human Resources

Challenge: The HR department has a lot of routine and monotonous work when the amount of employees is high.

Solution: NLP chatbots allow HR professionals to gather required information about employee performance, progress reviews, and other parameters. Chatbots are also effective during the recruitment process for scheduling interviews with prospective employees, answering prospective employees’ questions, and providing them with the necessary documents.

Sending updates and delivering notifications

Challenge: Dealers, distributors, and customers are not getting the latest information from your company.

Solution: With an NLP chatbot for any messenger application, you can automate updates and notifications for your dealers, distributors, and customers. The system can automatically decide when and what updates to send to each customer according to their orders, service ticket status, or other notifications.

Speech recognition without being informed which language is being spoken

Challenge: Presetting a translator and changing the settings when changing the target languages.

Solution: Since conventional speech recognition systems are developed separately for each language, users must select the language they want to speak beforehand. The accuracy of speech recognition systems also greatly suffers when dealing with overlapping speech by multiple speakers, thus limiting their applicability. As a result, usability is degraded due to the delay needed for language identification.

This new speech recognition technology has made it possible to operate devices such as smartphones and car navigation systems by voice. It simultaneously identifies and understands spoken languages without relying on expert knowledge such as phoneme systems and pronunciation lexicons.

Media monitoring

Technologies: NVIDIA DeepStream SDK, NVIDIA Riva, Triton Server, TensorRT, CUDA

Media monitoring system
Media monitoring system

Gather industry benchmarks

Challenge: Sometimes, manufacturers need to gather important information about the state of the market in a relatively short amount of time.

Solution: Web scraping can help to gather such valuable information as prices for raw materials and consumables and the cost of labor and services relevant to the business, as well as to get an idea of the state of the market in general.

Conduct compliance tracking

Challenge: Purpose-driven companies often include certain ethical production obligations in a supply agreement, concerning, for example, such issues as ethical workforce treatment. They want to be sure that their counterparts don’t break the agreement.

Solution: Web scraping can check publicly available information about suppliers and help to verify compliance with ethical practices.

Handle crises

Challenge: Since supply-demand balances depend on more than just the company’s internal processes, it’s highly important to become aware of external events that could impact supply chains as soon as possible.

Solution: Web scraping and media listening can be used to track critical external data, such as news about hurricanes, tornados, and other weather disasters, as well as local riots, governmental restrictions due to COVID‑19, and other events that could impact operations of suppliers and cause potential disruptions. Furthermore, algorithms can be used to scrape information from logistics carriers’ websites and help managers make decisions on whether to use different ways of shipment. Moreover, keeping track of a key supplier’s stock market status helps be sure of the supplier’s financial stability. Having this kind of information on hand, managers can take timely, preventive measures to mitigate risks.

Track potential epidemic field failures

Challenge: Receipt and usage of a defective batch of parts as a result of some manufacturing or engineering mistake might lead to a catastrophic situation that would result in a huge financial loss and a damaged reputation. The most famous example of a case of this is Samsung’s Note 7 smartphones bursting into flames.

Solution: In the case of mention tracking, it’s easy to notice when the number of mentions starts to grow unexpectedly. This tool provides the ability to detect crises when they first occur, and to take actions to mitigate further damage.

Feedback and ideas for improving products

Challenge: Lack of feedback or ideas on how to improve products.

Solution: Media monitoring allows for tracking mentions, detecting a lack of your product or service in a specific market, and testing out how people react to your product or service. It can also help you find the right people to influence the market, measure the success of your communication efforts, and help create campaigns.

Media Content Processing

Technologies: NVIDIA DeepStream SDK, NVIDIA Riva, Triton Server, TensorRT, CUDA

Media content processing system
Media content processing system

Automatic classification of incoming tickets

Challenge: It’s necessary to automatically categorize and process requests from various communication channels, such as social networks, email, and special systems, thus ensuring the proper level of customer satisfaction.

Solution: A classifier uses machine learning techniques to classify a ticket, then directs it to the most appropriate employee, and finally offers recommendations for solving the problem. The model is created and trained based on historical data related to successfully processed requests.

Threats and fraud detection by analyzing network traffic

Challenge: According to Sikich’s fifth Manufacturing & Distribution survey, half of the companies reported they had suffered a data breach or cyberattack involving their computer systems or networks over the previous 12 months.

Solution: NLP might be used to analyze network traffic and detect cyberattacks and threats, as well as other actions that may be considered fraudulent.

Reduce language barriers

Challenge: Manufacturing companies can’t pursue or may lose new business due to language barriers. Most organizations have global operations, and language barriers can hinder process efficiency.

Solution: Language barriers can limit the process efficiency of global operations. NLP can help with translating documents from one language to another without extensive human intervention, thus helping reduce regional language barriers. For example, it could translate customer-facing instruction manuals, customer service, call center materials, training and safety manuals, marketing materials, and equipment instructions.

Scene descriptions in natural language

Challenge: Factory automation systems can’t interact with humans using natural language.

Solution: The Attentional Multimodal Fusion technology is capable of weighing important unimodal information to support the appropriate choice of words for accurate scene descriptions. The system analyzes scenes to determine distinguishable visual cues and dynamic elements. The technology is expected to have widespread applications, including human-machine interface systems.

People analytics

Challenge: These days, when remote work is widespread, organizations may not always be aware of what actions their employees are taking.

Solution: NLP systems can analyze the digital footprints and changes in the digital activity of an employee in email and messengers through detecting thousands of data points, such as message content, response sentiment, and a communication graph. After that, the system can calculate the communication graph and predict the resignation trend of an employee before they decide to quit. Such systems can work at different sizes of manufacturing facilities and help them to keep their workers.

Recommendation systems

Technologies: NVIDIA Merlin, Riva, Triton Server, TensorRT, CUDA

NVIDIA Merlin recommendation engine
NVIDIA Merlin recommendation engine

Challenge: Workers are searching for information manually instead of having the system offering the right document at the right time.

Solution: Using the [people analytics approach], NLP systems can solve the problem of smart searches. For this, data such as letters, text files, messages in Slack, or other messengers are analyzed to understand what the employee is looking for. Thus, the system can predict what information an employee needs in the moment and make appropriate recommendations for the necessary data, or, for example, help find the right agreement or client contact data.

Document processing

Technologies: NVIDIA Riva, Triton Server, TensorRT, CUDA

Document processing pipeline
Document processing pipeline

Analysis of legal documents

Challenge: Any big business produces a large amount of various documents, including legal ones.

Solution: A system trained on a large number of legal documents will help in the analysis of new documents, identify risk factors, and suggest adding the necessary text inserts to those documents.

Analyze shipping documents

Challenge: A clear understanding of the current state of the supply chain is crucial to the smooth operation of a manufacturing pipeline. At the same time, every delivery, either domestic or international, must be accompanied by relevant documentation. Processing those documents is typically very time-consuming.

Solution: NLP can analyze thousands of shipment documents quickly and provide manufacturers with valuable information on what part of their supply chain is slowing down. With this information on hand, managers can make a well-considered decision on how to improve the process or what logistical changes to make.

Data entry task

Challenge: Data entry is a necessity that can be dull and time-consuming. Performing this task manually may result in data errors.

Solution: NLP-based algorithms understand both typed and handwritten text and can take over the job. They can extract data from large chunks of text and feed it straight into the database, saving time and energy for the specialists involved.

Reliability engineering

Challenge: Releasing a new product is a crucial time for a manufacturer. It’s extremely important to monitor clients’ complaints and returns in the early stages to get a clear idea of whether everything is going as expected or not. On the other hand, it’s quite difficult to process and classify all the service engineers’ reports (some of them written by hand), and as a result, a large part of them might be omitted.

Solution: The problem of having too many text reports needing to be read and classified sounds like a perfect task for NLP. A model is trained on a large number of reports labeled by quality engineers. Then, it can be fed the text from a report and give a prediction of the failure described in it. This way, significantly more products can be monitored, and all the huge backlog of reports can be automatically processed, thus allowing for drawing the right conclusions.

Automated report generation

Challenge: Report generation is often organized in such a way that managers draft them relying on the words of their subordinates. In cases like that, there may be a considerable loss of information from one level to the other. As a result, the prepared reports are not always as accurate as should be expected.

Solution: NLP can help to eliminate errors in reports. Machine learning algorithms can analyze the data stated by workers and form a report which then can be reconfirmed with them. A report generated this way is more credible and precise.

Knowledge processing

Challenge: The complexity of knowledge processing.

Solution: Knowledge processing is the systematization of such information as common human knowledge, experience, and sensations. They are systematized by a set of rules that promote mutual understanding between machines and people, as in the rule “if event A occurs, then event B effects.” Machines cannot accurately understand people’s intentions if they are not explained. When talking to another person, people tend to subtly verbalize their intentions, hoping that the other person will understand them.

The technology flexibly expresses various information in the form of a group of entities and relations. The group refers to a knowledge graph according to each situation. The knowledge processing also reduces the computational complexity of knowledge processing to approximately 1/3 to 1/10 of that of conventional processing.

Intelligent Search platform

Challenge: A recent study showed that workers in manufacturing waste about 76 hours per month searching for the information and combining/reformatting information from multiple sources.

Solution: As the distributed manufacturing world becomes increasingly digital, content and data are more and more emerging as a driving force for decision-making. The system increases productivity and minimizes time spent searching for information in unstructured data. It also helps with making better decisions, solving problems, eliminating the creation of duplicate parts, and many other things.

Text data Classification

Challenge: Large amounts of emails in inboxes, technical support, or service tickets demand a lot of time to classify them.

Solution: NLP helps to match text data with employees to save time and address each particular ticket to the most appropriate specialist.

Improve manufacturing deviation investigations

Challenge: To reveal trends, find patterns, and prevent deviation issues, it’s necessary to process a lot of documents.

Solution: An NLP system can look across large data sets and find correlations between obscure signals and events which the previous system could have missed. The system could show trends or point to more systemic issues, as well as explore weak signals that would be hard to find by human analysis. Data for machine learning takes from different sources, such as databases, CSV files, emails, documents, PDFs, and even binary data.

Collect knowledge needed for the repairing process

Challenge: It’s a matter of life and death to fix a machine that has malfunctioned, and even if it’s not dangerous, it’s always preferable to deal with a failure as quickly as possible. Maintenance technicians have to diagnose and fix issues without access to relevant cases or similar issues because the vast amount of critical data is unsearchable.

Solution: AI systems can help to reach the data from similar issues in the past and compare the information against technical guidelines.

Maintenance technicians fix the issue and enter their actions into the system to add to its knowledge. Maintenance managers can also identify trends of troubling issues in each season and take these insights to the original equipment manufacturers for improvement.

As a result of the analysis, it can be said that NLP technologies have been successfully applied in production for many years and have helped enterprises to move towards improving safety, efficiency, and productivity. Large companies are actively investing resources in the development of complex artificial intelligence systems containing NLP modules.

more posts

The ultimate guide to FPGA cards and Xilinx technologies
Anna Mosolova
Automatic speech recognition for specific domains
Anna Mosolova, Vladimir Nechaev, Marina Molchanova
Performance comparison of NVIDIA Riva 2.1.0 vs Riva 1.10.0
Vladimir Nechaev
NLP in manufacturing
Artemy Malkov, PhD
“Whose Ethics?”
Reid Blackman, Ph.D.
ML Models on Petabytes of Data — You Need GPUs
Paul Lashmet

Talk To Us

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
By submitting this form, you agree to our personal data policy