How to Choose an Industrial Monitoring or SCADA Implementer: Reuse the Existing System or Build a New One?
When a plant sets out to build industrial monitoring or a data dashboard, the usual disagreement is not whether to add new screens. It is this: when SCADA, PLCs or gateways are already in place, should the new platform reuse an upper-level interface, or build another collection path? Settle that before choosing an implementer, and the later discussion of protocols, acceptance and responsibilities becomes comparable.
Hongshan Tech delivers data integration, monitoring displays and custom systems on a per-project basis. Below are the boundaries to clarify during selection. Protocol testing and point-list checks still depend on site conditions and our existing technical guides.
On this page7 sections
First decide whether the project is monitoring or a replacement for site control
Classify an industrial monitoring project into one of at least three types before discussing the implementation plan:
- Monitoring display: equipment status, trends, energy use or output shown on a dashboard, board or digital twin for operators and managers.
- Alarm handling: beyond display, alarms are passed into an existing work-order or handling process, keeping their raised, cleared and acknowledged states.
- Replacing or intervening in site control: the new system writes equipment parameters, starts or stops equipment, or intervenes in interlocks.
Data access for the first two types should be designed as read-only monitoring. A protocol that supports writing does not mean the application should be given write permission. Our existing articles already state that data dashboards, operations boards and digital-twin displays should not, by default, replace on-site safety interlocks or safety instrumented functions. If the business really needs reverse control, it needs its own risk assessment and acceptance, and should not be bundled into the same scope as "getting the screens connected first".
If an implementer can only demonstrate screens refreshing, but cannot say whether control is involved, who approves writes, and whose instruction takes priority on site during a failure, the project boundary has not been made clear yet.
With SCADA in place, assess reuse first instead of reconnecting to the PLCs by default
When a plant already runs SCADA, a new monitoring page or management platform should first check whether the existing authorized interface or historian can cover the business questions this project needs to answer. If the fields, update rate, quality status and maintenance conditions are all suitable, reuse avoids duplicate collection. Only when key data is missing should other access paths be assessed together with the automation owner and the original vendor. A new platform being developed separately is not a reason to connect to the PLCs again by default.
Before comparing, draw the existing chain, for example: equipment / PLC, then the collection service, then SCADA, then the historian or interface, and finally the new platform. Remove any layer that does not exist. For each layer, record the maintainer, data scope, refresh or storage interval, conversion rules and access conditions. A trend curve may come from aggregated historian values, while a real-time state may exist only in the collection service; one cannot stand in for the other.
When comparing "reuse SCADA / historian" with "add device-side reads" for the same business task, check at least the following:
- Whether all required points are available, and whether units, scaling and state meanings are clear.
- Whether current values carry a source timestamp and quality status, and whether a stale value still shows after updates stop.
- Whether historical values are raw samples or aggregates such as averages or maximums.
- Whether alarms are independent events or calculated afterwards from values.
- Whether the interface is authorized for third-party use, and whether the vendor provides documentation and a test environment.
- Whether new reads add load on site, and whether they need separate access approval.
When only a few devices or fields are missing from the original system, a controlled hybrid path can be used: keep reusing what is available from the existing SCADA, collect only the missing items separately, and present them on the platform under stable equipment identifiers. Each field should have one responsible source, so the same output is never accumulated from two channels at once. When a source interface fails, the system should not quietly switch to another source while still showing "normal".
Choose protocols by their position in the chain, not by picking one of OPC UA, Modbus or MQTT
Our service descriptions note that near-real-time scenarios such as equipment status, energy use and alarms can be assessed for IoT, WebSocket, MQTT and similar access methods. Our industrial real-time data integration guide further explains that OPC UA, Modbus, MQTT and WebSocket often sit at different positions in the chain, and need not be treated as four mutually exclusive products.
A common combination is: Modbus reads registers from the device, a gateway forwards them to the platform over MQTT, and the platform pushes changes to the browser over WebSocket. Alternatively, an OPC UA client can subscribe to variables in SCADA or a device server. The combination should follow what the devices actually support, data semantics, network zones, reliability and security requirements, and be confirmed by integration and testing, not promised as "supports everything" based on marketing claims.
What matters more than reciting protocol names is whether the implementer can explain, based on the existing system, which layer to use, which points to read, who authorizes access and how acceptance works. Whether a given product implements a protocol's subscription, event or security features still has to be checked against the actual model and configuration. Business meaning is not unified just because the protocol name is the same.
Without formal interface documentation, research and a small amount of read-only verification can come first, but unknowns should go on a risk list rather than being promised as full integration scope. Legacy equipment need not be assumed to require wholesale replacement either. Check existing interfaces, protocols and available signals first, verify the collection path and data meaning on representative devices, and then decide on adaptation, an added gateway or a limited retrofit.
Acceptance checks points, timestamps, quality and disconnection, not how good the demo looks
Demo or sample data can be used to confirm page layout and interaction, but a changing number on a page does not prove that a whole class of devices is connected. Interfaces that are not yet connected should be marked as unverified, and should not be accepted as a live real-time system.
For key points, check at least the following:
- Point identity and meaning: code, owning equipment, unit, data type, scaling and read/write attributes, cross-checked against on-site values.
- Timestamps: distinguish device event time, gateway collection time, platform receipt time and storage time; agree which one drives trends and alarms, and record the time zone and clock source. The display layer must not guess times.
- Data quality: the application should recognize states such as normal, invalid, offline, out of range, manually substituted or unknown source, rather than substituting zero for missing data.
- Disconnection and recovery: the system recognizes offline states and does not present the last good value as a current real-time value; after recovery, check the backfill range, duplicates, ordering and gaps. A restored network does not mean the data is complete.
- Permissions and audit: who can read, whether writing is allowed, how test accounts are revoked, and whether configuration and permission changes are traceable.
Write collection frequency and end-to-end latency separately. The first answers how often the source produces or reads data; the second answers how long a change that has already happened takes to appear on the target page. Accepting only a page refresh animation cannot prove the source data is fresh enough. Specific frequencies, latencies and recovery targets must be agreed in writing based on site conditions and business consequences, not taken from generic numbers detached from the project.
If the monitoring screen runs on an unattended terminal, also check the stale-data warning after a network loss, identity and state verification after reconnecting, and whether the screen returns to a usable view after the browser or the whole machine restarts. A fixed terminal should restore only the agreed page, never unconfirmed control actions; whether the collection side backfills data from the outage should not be guessed by the browser.
An implementer should explain the division of responsibilities, not just promise "we can do it all"
Equipment monitoring is not simply gateways plus sensors plus a set of pages. Signal sources, collection, network, storage, application and site work should each be listed with inputs, deliverables, owners and acceptance evidence. Each party separately confirming that "our product works" does not prove the whole chain is connected.
In early conversations, rather than asking "can you do the whole thing", you can provide: existing system or equipment models, known interfaces, the states you need to see, the user roles, the deployment environment (such as an intranet or private deployment), and how far the vendor and site can cooperate. Mark anything you cannot provide as to be investigated. The implementer should then list separately what can be confirmed from existing information, what needs verification, and what must be supplied by hardware, network, vendor or other parties.
Within the service boundaries Hongshan Tech has published, data platform and visualization work can cover interface adaptation, status display and record queries around confirmed data sources. Hardware supply, electrical modification, site construction and control modification are not automatically included in the software scope. Network provisioning, vendor licenses, test windows and point changes also need to be assigned in advance.
Intranet or private deployment can be assessed, and data does not necessarily have to go to the public internet. But that is a deployment boundary, not inherent security. Network zoning, identity and permissions, transport protection, log auditing and backup and recovery still need to be confirmed according to project risk. Whether graded protection (classified cybersecurity protection) or industry-specific requirements apply should be judged by the owner and security professionals.
For project delivery, the process published on our homepage has seven steps: requirements discussion, solution design, prototype confirmation, development and implementation, testing and acceptance, deployment and launch, and operations support. For industrial monitoring projects, the requirements stage should likewise confirm whether to reuse the existing system or add collection, and whether the scope is read-only monitoring. The testing and acceptance stage should keep evidence against the point list, timestamps, quality status and disconnection recovery, rather than judging only the pages in a demo environment.
FAQ
If a plant already has SCADA, should it still connect directly to the PLCs?
Not necessarily. First assess whether the existing authorized interface or historian can serve this business task. If fields, frequency and quality status are all suitable, reuse comes first; if there are real gaps, assess other access paths with the automation owner and the vendor. A separately developed platform is not a reason to reconnect to the PLCs by default.
Can an industrial monitoring dashboard send control commands directly?
Monitoring display and alarm handling should be designed as read-only. A protocol that can write does not mean the application should get write permission. When reverse control is needed, assess permissions, confirmation steps, failure behavior and the relationship with site interlocks separately, rather than folding it into screen integration as a default deliverable. A display system must not replace on-site safety interlocks or safety instrumented functions.
What does integration acceptance check? Does a page showing refreshing data count as a pass?
No. Check point identities, units, timestamps, quality status, update timeliness, disconnection recovery and permission records. Demos or samples can confirm layout, but unverified interfaces cannot be accepted as a real-time system.
Must the data stay on the intranet? Is an intranet deployment automatically secure?
A data dashboard can be assessed for deployment on the organization's own servers and intranet, subject to actual testing in the target environment. Intranet or private deployment is only a boundary condition; permissions, auditing, backups and operations access still need to be designed. "Deployed on the intranet" should not be treated as "already secure".
What should you clarify first when choosing an implementer?
Clarify whether the project is monitoring, alarm handling or involves control; which layer of the existing system will be reused; which points and which time, quality and disconnection records acceptance will check; and who is responsible for software, hardware, network, vendor and site construction. If these boundaries cannot be explained, the decision should not rest on demo effects alone.
Related reading
- Connecting a Data Platform to an Existing SCADA: Reuse an API or Read the PLC?
- Industrial Real-Time Data Integration and Acceptance: PLC, SCADA, MES, and IoT
- Connecting Legacy Equipment: Interface Assessment and Minimal Retrofit Options
- Equipment Monitoring Delivery Boundaries: Acquisition, Network, Software and Site Work
- Recovering an Unattended Dashboard After a Connection Failure
- What to Prepare for Private Deployment of a Data Dashboard
This article explains a method for project selection and acceptance. It does not replace a site survey, and it is not a unified protocol support list or a delivery commitment. The combined paths and gap handling described are method examples, not delivered client projects.